Visualization of training data reportedby football players

Full text

(1)Bachelor of Science in Computer Science May 2018. Visualization of training data reported by football players. Olof Christensson Adam Georgsson. Faculty of Computing, Blekinge Institute of Technology, 371 79 Karlskrona, Sweden.

(2) This thesis is submitted to the Faculty of Computing at Blekinge Institute of Technology in partial fulfilment of the requirements for the degree of Bachelor of Science in Computer Science. The thesis is equivalent to 20 weeks of full time studies.. The authors declare that they are the sole authors of this thesis and that they have not used any sources other than those listed in the bibliography and identified as references. They further declare that they have not submitted this thesis at any other institution to obtain a degree.. Contact Information: Author(s): Olof Christensson E-mail: olle.ch@hotmail.com Adam Georgsson E-mail: adamgeorgsson@gmail.com. University advisor: Assistant Professor Prashant Goswami Department of Creative Technologies (DIKR). Faculty of Computing Blekinge Institute of Technology SE–371 79 Karlskrona, Sweden. Internet : www.bth.se Phone : +46 455 38 50 00 Fax : +46 455 38 50 57.

(3) Abstract Background. Data from training sessions is gathered by a trainer from the players with the goal of analyzing and getting an overview of how the team is performing. The collected data is represented in tabular form, and over time the effort to interpret it becomes more demanding. Objectives. This thesis’ goal is to find out if there is a solution where collecting, processing and representing training data from football players can ease and improve the trainer’s analysis of the team. Methods. A dataset is received from a football trainer, and it contains information about training sessions for his team of football players. The dataset is used to find a suitable method and visualize the data. Feedback from the trainer is used to determine what works and what does not. Furthermore, a survey with examples of visualization is given to the players and the trainer to get an understanding of how the selected charts are interpreted. Results. Representing the attributes of most importance from received dataset requires a chain of views (usage flow) to be introduced, from primary view to quaternary view. Each step in the chain tightens the level of details represented. Box plot proved to be an appropriate choice to provide an overview of the team’s training data. Conclusions. Visualizing training data gives a significant advantage to the trainer regarding team analysis. With box plotting will the trainer get an overview of the team and can hereafter dig into more detailed data while interacting with the charts.. Keywords: Visualization, charts, statistics, training-data,.

(4)

(5) Acknowledgments Big thanks to: Prashant Goswami: who led us through the thesis process Herman Ottosson: for the idea behind the thesis and the feedback Martin Söderberg: for useful sport input. iii.

(6)

(7) Contents. Abstract. i. Acknowledgments. iii. 1 Introduction. 1. 2 Related Work. 3. 3 Method 3.1 Implement web application . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Player survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 5 7 10. 4 Results 4.1 Web application . . 4.1.1 Main views 4.1.2 Other views 4.2 Survey . . . . . . .. 13 13 16 19 20. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. 5 Analysis and Discussion. 23. 6 Conclusions and Future Work. 25. References. 27. A Supplemental Information. 29. v.

(8)

(9) List of Figures. 3.1. 3.2. 4.1. 4.2. 4.3. 4.4. 4.5. 4.6. 4.7. Dr Abela’s chart chooser [6]. The image was made to help the choice of the selecting fitting charts depending on what type of data are presented. It is a flowchart that begins in the middle and with the help of questions leads to the right chart for the task. . . . . . . . . . . . . . Borg scale for estimated effort [1]. The perceived exertion in the body can be described with a number from a scale 0-10 where 0 is rest and 10 is maximal exertion. . . . . . . . . . . . . . . . . . . . . . . . . . . Example of stacked bar chart for the attribute effort. The bars on the y-axis are only populated with players that are outside the expected effort interval. The blue bar is displaying ±1, yellow ±2, and orange <=±3 from expected effort interval. The label displayed on the bars are representing the number of entries inside the bar. . . . . . . . . . Example of how bubbles in bubble chart are expanding over each other for the attribute effort. Each group of colored-bubble represents one weekday from Monday to Friday with Wednesday removed. All y-axis values for the green bubbles are 3,4 and 5 which can be hard to see due to the width of the bubble. . . . . . . . . . . . . . . . . . . . . . . . . Example of box plot for the attribute effort. Y-axis displays the effort for the players. Each colored-box represent one weekday from Monday to Friday with Wednesday removed. . . . . . . . . . . . . . . . . . . . The flow between each chart on the web application. Each step presents a new view with more precise data with which the user can interact. Except for the final view when the period and data type is expanded again. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Box plot combined with table for effort. The box plot belongs to the primary view and has been interacted with so the secondary view with the table has been unveiled. . . . . . . . . . . . . . . . . . . . . . . . . The tertiary view is a pop-up for the effort which discloses when a player name is clicked on the secondary view. The chart in the tertiary view is a line chart with a single line symbolizing the chosen played and how it relates to the expected effort area that is greyed out. . . . . An extraction from the quaternary view for effort (the blue line). This chart is similar to the tertiary view but with an extended time period. It is also combined with freshness (the green line). Injuries and extra training sessions are marked as red and orange dots. . . . . . . . . . . vii. 6. 9. 14. 15. 15. 16. 17. 18. 18.

(10) 4.8. 5.1. Charts used in the survey. Top left: stacked bar, top right: table form, bottom: box plot combined with a table. Five questions about how easy the charts are to read where asked for each chart. Box plot was also asked about individually. . . . . . . . . . . . . . . . . . . . . . . . . .. 20. An unreadable example of line chart representing 15 players’ reported effort. The reported effort is for one week. The chart was made during implementation of the web application. . . . . . . . . . . . . . . . . .. 24. viii.

(11) List of Tables. 3.1. Attributes that the players are reporting after every training session. .. 4.1. Languages, libraries and their version numbers used to implement the website. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mean values for each question, collected from the result of the survey done by players. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Answers for each question, collected from the result of the survey done by the trainer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 4.2 4.3 A.1 A.2 A.3 A.4. Survey Survey Survey Survey. result result result result. for for for for. box plot . . . . . . stacked barchart . table . . . . . . . combination of box. ix. . . . . . . . . . . . . . . . plot and. . . . . . . . . . table. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. 8 13 21 21 37 38 38 39.

(12)

(13) Chapter 1. Introduction. With the help of available technical assistances, the majority of sports clubs and individual athletes with a serious goal are recording and analyzing their activities to maximize the resulting outcome. For a single athlete, it is relatively easy to track her training with the wide range of tools and applications available, but applications targeting a whole team are not as many. The majority of existing applications seem to mainly focus on the sport itself (training schedule, exercises, game session, match result etc.) and not the athletes within the team [9, 10]. Feedback from the athletes can be a crucial part of how the training is scheduled. A single trainer of a sports team usually has an expected view and plan of how each training session should look like, to in the end reach a specific goal. Therefore it is of importance to the trainer that each athlete in the team is performing as expected on the sessions. For an athlete to achieve the proper results from each training, they are often requested to eat, sleep and drink (stay hydrated) enough to be able to perform. To manually keep track of each athlete status for example, how well they are performing, injuries, sleep, can be both tiresome and time-consuming. Over time it might get harder to keep track of the history of each player. Also, it is difficult to keep track of multiple players and follow each player. If one or more players are not living up the expectations, the trainer needs to figure out why. Data presented in charts is easy to grasp, and it is a quick way to familiarize with it, but it is essential to find the right way to present the data [17]. Football teams consist of around 20 players, and this could limit what kind of charts that are suitable. There are plenty of ways representing data, and it is often possible to combine them to get a better result. One way to do this is an interactive presentation where it is possible to show additional information by interacting with each data entry. In an article by Albinsson and Andersson, they show how analyzing football statistics can be improved by combining different datasets [15]. All variables from a dataset might not fit in one single chart, but splitting them apart in two and linking them together could be a possible option. The second chart behaves differently depending on what interaction occurs in the first chart. Too many variables and details in a chart make it troublesome to read. Dividing the chart into several charts may help, but a significant amount of charts can be overwhelming to the user if not presented carefully. When exploring the visualized data the user often needs an overview of the charts and a dashboard could help with that. A dashboard "is a visual display of the most important information needed to achieve one or more objectives consolidated on a single screen so it can be monitored and understood at a glance" [11]. 1.

(14) 2. Chapter 1. Introduction. This thesis’ goal is to find out if there is a solution where collecting, processing and representing training data from football players can ease and improve the trainer’s analysis of the team. The dataset that will be used is received from a trainer with the requirement of providing a better and more accessible overview of the content of the dataset. The data has been collected using a form, (Google Form) [2], where the players have been encouraged to fill in after each training session. The form offers the possibilities to visualize the answers as bar charts which are processed from the entire dataset. The dataset is continuously growing, and the bar chart ignores the dates which prevent a periodic overview. Furthermore, Google Form offers the option to export the collected data as a spreadsheet, but analyzing a large spreadsheet can be both tiresome and time-consuming. To provide a better overview and to ease the analysis for the trainer, a more robust solution is needed. The research question this thesis seeks an answer for is: To what extent can visualization ease the analysis of training data compared to presenting it in tables? By developing an application which can gather and represent the data from the training sessions for the training, it can ease the analysis process. The application should suitably visualize the data so the trainer easily can get an overview of the team. With the help of the application, the trainer can locate outliers and more focus can be put on the players behind the abnormal data. An abnormality could originate from the fact that players have not understood what the trainer expects of them on a specific training session. It could also arise that the trainer has put too many high-intensity sessions after each other and not allowing the players to rest enough for the next training. By moving the gathering of data to an application, the application could help to solve the problem with missing reports since functionality to remind the players to fill in the form can be implemented..

(15) Chapter 2. Related Work. Several studies have been done on how to help sports teams continue to improve, both as a team and the individual athletes within the group. The studies stretch from targeting individual athletes and how they can improve the team-spirit itself. They also often concern the issue "What can the athlete/team change or improve to get one step closer to the goal?" One approach is a so-called semi-automated monitoring system by using video surveillance, i.e., Motion analysis [7, 16]. A number of cameras are used to capture the training or game session and can be analyzed afterward. By using a semi-automated system, the dependency on the individual to self-estimate after the training is removed. Despite the advantages of these semi-automated systems, they do not provide the big picture. They can only offer what can be seen on the tape and not the reason behind the captured performance for the session. Different surveillance has different disadvantages, but a few common are cost, need for right lighting conditions, the setup of multiple cameras and most crucially understanding the captured video [16]. Similar work has been done as the video surveillance but with Global Positioning System (GPS) where the authors tried to find a correlation between the position of a football player (defender, midfield, forward, etc.) and the distance covered during a football game. With the help of the GPS, they concluded that the distance traveled during one football game are 9900 ±700 meters [8]. It was also found that each position differs by distance traveled. For example, the defender was discovered to have shortest distance traveled and the least amount of short sprints, but furthest distance in the lowest speed range. That seems to be the opposite to the other position midfield (outer-midfield) which had the longest distance traveled in the highest speed range, but the shortest distance traveled in the lowest range. Another study which also used GPS to find a correlation of distance traveled shows similar result [12]. The results from both of the GPS-studies indicates that football trainers need to adapt the training depending on consideration of what positions the players have. An application that can complement motion analysis or used as standalone is a self-reporting system, for example, Trimbite [5]. By allowing each athlete report and provide their view of the training or game session could help the trainer to understand the squad better. Trimbite has been used by a Swedish football team to support the trainer monitor the squad current status [3]. According to the newspaper, by reporting training load and recovery using the application contributed to a reduced number of injuries for the team. Analyzing reported training data is often done by some sort of visualization. It is not always the best option though to visualize data 3.

(16) 4. Chapter 2. Related Work. instead of keeping it in tables. M. McBride et al. are in [13] investigating how tables compare to graphs when an agent reads network traffic from dark networks. The agents’ task is to make decisions on how to disrupt a network based on the network traffic. They conclude that for their particular needs are graphs better to make decisions quickly, but weakly better disruptions emerge from reading tables. Charts can also be misunderstood if not constructed correctly, as Dr. Abela writes in [6] In [14] L. Nuzzo mentions that bar charts can be used for summarizing counts or proportions with categorical data, but might not be the most optional to compare numeric responses. Furthermore, the bar can be misleading depending on how the length/width and start point are visualized which were taken into account when the bar charts were drawn. For example, the reader might expect the bar to always start at zero which can cause a stacked bar chart to be hard to understand. The author also mentions an alternative chart to be used instead for visualizing measurable data: box plot. Box plot uses statistical summaries (median and interquartile range) that are more robust than bar chart when there is a lack of data. Compared to a bar chart, box plot has been implemented to take outliers into account by highlighting them. It firstly draws the box plot without the outliers and secondly places the outlying values as single dots outside the original box, which eases the effort to find them. The article mentions a few drawbacks of using box plot, and it replaces the population available with percentage making it impossible to see how many entries the chart contains. Also, compared to bar chart is box plot not a well-known chart..

(17) Chapter 3. Method. To answer the research question: To what extent can visualization ease the analysis of training data compared to presenting it in tables?, a selection of charts are chosen to visualize the dataset. The choice of chart depends on what the data looks like, what the intention to get from it is and feedback from the trainer. Each chart has its limits on the number of variables and datasets it can contain and which data types it can represent. Each chart type also has its strengths, and some chart types are better at showing the relationship between data for example. Dataset The dataset received consists of collected attributes (shown in table 3.1) from 20-25 male football players during five months. The players belong to the same team. The age span of involved players are 15-30, and the occupation of the players are unknown. All the players are encouraged to fill in the form created by the trainer, even on a day without training. There is no verification to assure all players are answering the form and the risk for unrecorded data is high. Chart capacity All collected data reported includes a corresponding date of when it occurred and to visualize how information changes from session to session, the date will be used on the x-axis. To visualize how the data varies between training sessions, a chart with the capacity of comparing the entries was necessary. Figure 3.1 was used to determine a suitable chart and the figure proposed to use bar chart or line diagram, the most straightforward chart which for the majority of people find easy to understand. Depending on the number of visible lines in a line chart, it could cause the represented data to be troublesome to understand. Furthermore, there are not many variants of the line chart. However, when there are only one or few entries per x and y-axis, the line chart is more suitable. Compared to lines, a bar chart can be done in multiple different ways. For example wide bars, stacked bars, waterfall bars, negative and positive bars. Continuing were bubble chart for relationship data tested, the larger the bubbles are, the more entries on the y-axis which provides excellent visualization of reported effort and spread. Drawbacks on bubble chart are that multiple different entries on the y-axis could hide other bubbles when expanding and the scale of the bubbles could vary depending on the number of reported players. The visualized chart needs to be represented similarly each time it is drawn to ease the interpreting of the data 5.

(18) 6. Chapter 3. Method. Figure 3.1: Dr Abela’s chart chooser [6]. The image was made to help the choice of the selecting fitting charts depending on what type of data are presented. It is a flowchart that begins in the middle and with the help of questions leads to the right chart for the task..

(19) 3.1. Implement web application. 7. shown. The variable that causes the same variant of a chart to vary is the number of participating players and ignoring it could ease the process of understanding the charts. One chart which is not present in figure 3.1 but could be placed next to bubble chart is box plot. As mentioned in chapter 2, using box plot removes the possibility to see the participants for the visualized diagram. The sample for the box plot is the number of participated players for each training day. For a trainer it might be interesting to know the sample of the displayed chart and if box plot itself is unable to visualize that, it can be enhanced with an additional chart. The disadvantage of visualizing with charts is the level of detailed data that can be displayed and the risk of misunderstanding the visualized data increases [6]. Compared to charts, representing the data in tables can practically present the data in its raw form but in a structured way. With tables comes the possibility to sort and filter the represented data which makes locating specific data points more obtainable. One drawback with tables is that they can rapidly develop into a significant number of columns and rows. This could create an overwhelming content of data and getting a feasible overview becomes challenging. Feedback was received from the trainer to include expected effort range on the overview chart for effort. Including the span can provide a quicker analyze if the team is within the expected effort range, which is achievable by complementing box plot with two lines (upper and lower expected effort). Two methods will be used to find out how visualization can ease the analysis of training data. One of them is an implementation of a web application aimed to address football trainers. As a second method, a survey will be given to the trainer and his team to back up the decision of charts used in the web application.. 3.1. Implement web application. To efficiently investigate how different data visualizations performs is web application a suitable choice due to the vast amount of available languages and libraries. No matter the selection of libraries or languages most devices or platforms can handle a web application, all that is needed is an internet connection. The data collection can be done by a simple HTML-form where players report their data which are submitted to a database. Several scripting languages such as JavaScript, Python and PHP are suitable for web development. Since all chart types often demand their own kind of formatted data, the scripting language will perform the processing. Choosing which type to use and how to design it is the crucial part in this thesis. Box plot, bubble chart, pie charts, tables, line charts and several kinds of bar charts have all been tested and used during the development. Feedback from the trainer has been used on how well the charts convey their underlying data. The form the players will fill in is the same as the trainer previously has used, but several of the questions have been changed slightly to let the answers be converted to integer values. The conversion into integers eases the processing and comparing of the data values. Three of the questions are free text (anything can be written) and will not be processed and visualized. Instead, the answers will be listed in tables or used as help text for hover events..

(20) 8. Chapter 3. Method. Reported attributes In table 3.1 are the attributes listed that the players are expected to report via a form every day even if no training session has occurred. Attribute Training date to fill in for Effort Freshness Self-estimated performance Sleep hours Number of cooked meals Number of snacks Breakfast Urine color Extra training Injuries Other comments. Possible answers Today’s date and 30 days back 0-10 1-5 4-11 0-3 Yes, No Dark, medium, light, transparent Free text. Table 3.1: Attributes that the players are reporting after every training session. The collected attributes of most importance for the trainer are effort and freshness. The effort is how exhausting a player experienced the training was on the Borg scale, 0 to 10 where 0 is rest, and 10 is the maximal effort (figure 3.2). Each training session has an expected range on the Borg scale wherein the participating players are expected to be within. Expected effort varies from day to day, and an effort value of 3 can be within range one day but not the next day. The trainer is interested in the number of players that are outside the expected effort-range. Freshness is the "body feeling" the players have at the end of the training session (or day if no training occurred). The range of freshness reaches from 1 (very fresh) to 5 (very worn down). Players can be expected within the upper freshness range (4 or 5) if the training session had a high expected effort-range, but a lower range (1 or 2) is not necessarily bad. Compared to the effort, freshness does not have an expected range. The remaining attributes (sleep, breakfast, meals, self-reported performance and urine color) will be summarized and evaluated together except attributes with free text. To be able to achieve that, a point system will be created where a single player receives different points depending on reported values. Alarming values will be evaluated to a low final score and vice versa for non-alarming values. All points added to a total correspond to a final score (Poor, Bad, OK, Good-performing). The final scores could provide an indication of how acceptable values each player has reported during the week. Usage flow When representing the values from the attributes effort and freshness, one requirement from the trainer was to provide an overview of how the values are divided.

(21) 3.1. Implement web application. 9. Figure 3.2: Borg scale for estimated effort [1]. The perceived exertion in the body can be described with a number from a scale 0-10 where 0 is rest and 10 is maximal exertion..

(22) 10. Chapter 3. Method. between the players. For effort, it was important to highlight any outliers. Outliers for the effort is a player who has reported an effort of for example 9 when the majority of the other players are within 2-5. Another requirement was to have the possibility to quickly look into what the reported values were for a chosen day from the overview chart. To meet the wanted requirements a usage flow needs to be introduced where the trainer uses a series of clicks or hover events (hover over part of the chart with the mouse to display additional information). The usage flow provides a path between charts to enable additional information for each step in the chain. With the usage flow, the charts can be made more straightforward and easier to understand.. 3.2. Player survey. The target group for the survey consists of the football team’s players since they already are familiar with the term effort. The survey will indicate how easy or hard the players think various chart types are to interpret. Charts from the implementation were used as models in the investigation. Further on, the same questions are sent to the trainer so a comparison of how conversant individual answers differ from the team’s. The result will be used for the application to back up the decisions on which chart types to use. The dataset from the trainer was not used for this survey due to the lack of data, a minimum of 15 players was decided to be needed. Therefore a fake dataset was created with 15 players for one week. Each day during the week have at least 12 data entries and for every day there exists at least one player who is way off on the effort scale. Even if the data is simulated, the players will still recognize it as they have reported and worked with similar data before. That way futile effort will not be put to understand what the underlying data contains. Since effort is the primary category for the trainer, charts in the survey only included data points from that category. The question set is about what characteristics the trainer is looking for in his players’ data. The aspects of the items are about finding outliers, number of players (both as a hole and on each value), the spread and the proportion. Before the survey is sent out to the players, it will be tested on a few colleagues to detect if the diagrams are hard to understand or if the formulations of the questions are poorly. Each question has the same possible answers, a scale from one to five, where one is very easy, and five is very difficult or even impossible. 1. How easy/difficult do you think it is to see how many players attended the Friday training? 2. How easy/difficult do you think it is to see which day was most exhausting for the team? 3. How easy/difficult do you think it is to see how broad the spread among the players is? 4. How easy/difficult do you think it is to detect an outlier? Outliers are players who differ from the rest of the team..

(23) 3.2. Player survey. 11. 5. How easy/difficult do you think it is to detect how many had an effort of 4 on the Friday training?.

(24)

(25) Chapter 4. Results. 4.1. Web application. Implementation of a web application was successfully achieved. On the site, players can report their data, and the trainer can obtain the result from different charts divided into parts of a specific usage flow. To visualize the charts on the website the JavaScript library Plotly JS was used due to the well-documented reference page [4]. Remaining parts of the site was implemented using PHP, HTML/CSS, BootStrap, JavaScript and MySQL as the database. Library/language Plotly JS PHP HTML MySQL BootStrap. Version number ^v4.0 ^v5.4 ^v5.0 ^v4.x ^v4.0. Table 4.1: Languages, libraries and their version numbers used to implement the website. Selection of charts It was expeditiously discovered that the tested comparison charts (line chart and several combinations of the bar chart) were more suited for describing a single player compared to the whole team. The comparison charts rapidly became skewed and challenging to understand. Several variations of bar charts were tested with the yaxis representing either amount of players or attribute level. Color coding and labels were added to complement the bars further. When a chart visualizes effort for the team and y-axis represent the number of players, the bar colors indicate on what effort level the bar symbolizes. In a worst-case scenario could the number of bars for a single day exceed what is considered readable, but stacking the bars on top of each other solves that. Therefore did stacked bar chart turn out to be the best candidate of the bar charts for the provided dataset. In figure 4.1 is the y-axis representing the total amount of players while the labels inside the bars are indicating how many players reported that specific value. The colors show how much the players differ from the expected effort and bars that goes below the baseline represents a negative difference. This chart succeeds to express what is sought, but the reader is exposed 13.

(26) 14. Chapter 4. Results. to too many details for the chart to be a viable option when it comes to giving an overview that is quick to grasp.. Figure 4.1: Example of stacked bar chart for the attribute effort. The bars on the yaxis are only populated with players that are outside the expected effort interval. The blue bar is displaying ±1, yellow ±2, and orange <=±3 from expected effort interval. The label displayed on the bars are representing the number of entries inside the bar.. Bubble chart and box plot are better suited for relationships as partly seen in Dr. Abela’s chart chooser in figure 3.1. For the used dataset, bubble chart was shown not suitable due to it occurred too often that the drawn bubbles expanded over each other (see figure 4.2). A smooth transition for the bubble between the scale of for one player and the scale for multiple players (10-20) was not found, due to an irregular number of players for each day. A large scale could cause the circle to expand over the others, and a small scale could make the outliers hard to identify due to its size. Therefore is bubble chart not a good choice when the number of participated players varies. The chart was also proven to be unsuited for the usage flow due to problem of identifying which bubble the click event was triggered on. A chart which ignores the population was needed and box plot proved to do that (figure 4.3). Regardless how ranging in size the datasets are, box plot keeps its layout. Also, feedback from the trainer revealed that box plot was a desirable way to represent the different attributes..

(27) 4.1. Web application. 15. Figure 4.2: Example of how bubbles in bubble chart are expanding over each other for the attribute effort. Each group of colored-bubble represents one weekday from Monday to Friday with Wednesday removed. All y-axis values for the green bubbles are 3,4 and 5 which can be hard to see due to the width of the bubble.. Figure 4.3: Example of box plot for the attribute effort. Y-axis displays the effort for the players. Each colored-box represent one weekday from Monday to Friday with Wednesday removed..

(28) 16. Chapter 4. Results. Usage flow Each step in the usage flow, called view, will have its chart starting from primary view down to quaternary view as shown in figure 4.4. Each chart will provide a weekly overview with the possibility to jump between weeks. From the primary view, the trainer can continue further down the usage flow using a series of clicks. For each click, another view is displayed, and for each view, the displayed data becomes more specific and more detailed until the user has reached the quaternary view. As the initial goal of the web application was to provide an understandable overview of the essential attributes (effort and freshness), the primary charts will contain those two.. Figure 4.4: The flow between each chart on the web application. Each step presents a new view with more precise data with which the user can interact. Except for the final view when the period and data type is expanded again.. 4.1.1. Main views. Primary view Box plot was selected to act as the start point in the usage flow for both of the attributes (effort and freshness) due to its ability to provide a good overview and highlight outliers. The box plot for the effort was enhanced with two lines displaying the interval which the players are expected to be within. The area between the lines is colored to indicate where the box is supposed to lie. Each drawn box have a click event attached, and will trigger the secondary view to show when clicked. There is no static expected freshness for the players. There is, however, an anticipated trend of how the freshness will change over the week. After an exhausting training session, it is not uncommon that several of the players are reporting high (4 or 5) freshness..

(29) 4.1. Web application. 17. Figure 4.5: Box plot combined with table for effort. The box plot belongs to the primary view and has been interacted with so the secondary view with the table has been unveiled. Secondary view A tabular form was selected for secondary view due to its ability to provide useful insight into specific data entries and to complement the primary view. The drawn table will contain values which box plot from the primary view was unable to provide, i.e., the total number of players and data values for the clicked box that day. Players who are within the expected effort are by default hidden under a label (an HTML tag). The hidden athletes can be displayed by manually clicking the blue marked label Click here to display x more players inside the expected effort range. Each player label will have another click event connected to them and when clicked will trigger the tertiary view to be displayed. Tertiary view Line diagram was selected for the tertiary view to display attributes for one single player over time. The presented data for the line diagram is for the chosen week, but it has the feature to scroll back and forth in time quickly. Similar to primary view, the tertiary view was enhanced with two lines displaying the interval of expected effort. The tertiary view is presented as a pop-up window (using modal in HTML), removing the need of navigating to a new page. The pop-up will hover above the primary and secondary view when triggered, see figure 4.6. The tertiary view for freshness is similar except the expected interval lines. From the pop-up, the trainer can choose to go back to the secondary view or continue to the quaternary view. Quaternary view To provide the possibility to view all data categories for one single athlete a quaternary view was created. Here are all data entries represented in different charts. For a single player, there will only be one entry per type of data. Line chart and bar chart is used to describe the entries over time and the free text fields are listed with.

(30) 18. Chapter 4. Results. associated date. See figure 4.7 for an example.. Figure 4.6: The tertiary view is a pop-up for the effort which discloses when a player name is clicked on the secondary view. The chart in the tertiary view is a line chart with a single line symbolizing the chosen played and how it relates to the expected effort area that is greyed out.. Figure 4.7: An extraction from the quaternary view for effort (the blue line). This chart is similar to the tertiary view but with an extended time period. It is also combined with freshness (the green line). Injuries and extra training sessions are marked as red and orange dots..

(31) 4.1. Web application. 4.1.2. 19. Other views. In addition to the main views, there are also other views which do not have the same priority for the trainer but are still important. Comments and injuries Since the free text attributes entries are hard to interpret were they not visualized. Instead the written text received from the form will be represented in tabular form with weekly intervals, ordered by weekdays. Each entry in the table will contain the date when the free text was submitted, the content itself and name of the author (player). Clicking on a player’s name leads directly to the quaternary view for that player. Players overview and latest freshness/injury The trainer asked for a page where the possibility to see the latest reported status for all players. After further discussion, it was concluded that only latest freshness and injury was wanted. As a result, an overview-page of all players present in the team was created. On the page, it is possible to view most recent freshness and injury. Since it is only one entry point per player, a table form was the best way to represent this kind of data. Point system To provide a better overview of the collected data from the players, including all data entries, a point system was developed. The system summarizes all the recorded attributes for one day (except the free text) and calculates points depending on the reported value for the categories. The number of reported training sessions thereafter divides the computed value. Four different intervals are used to determine a result where the summarized value lies and how good the value is, displayed in the bullet point list below. The calculated result presents an overview for the trainer, describing if the outcome for each player is negative or positive. The trainer decides what values there are to be considered positive contra negative. Negative values on several players or a single player should not be reviewed as a grade or final score but act as a flag or indicator for the trainer. The indicator should hint the trainer that one or more players are not reporting the expected values, for example, less than six hours of sleep or only one meal that day. The point system is visualized in tabular form combined with a progress bar displaying how the summarized value relates to the maximum amount (100). From the point system, there is a navigation link to the individual player views provided. The four different intervals are: • Above 90: Excellent performing • Between 90 and 70: Good performing • Between 70 and 50: OK performing. • Below 50: Bad performing..

(32) 20. 4.2. Chapter 4. Results. Survey. The resulting charts from the implementation were used in the survey and are shown in figure 4.8. Three different charts are represented in the survey: boxplot, stacked bar chart and a table (all charts used the same dataset). There was also a combination of two charts to see if it brings complementing advantages. In the combination is the box plot used again next to a simplified version of the table. It simulates that a box has been selected and the table only displays the data belonging to that box.. Figure 4.8: Charts used in the survey. Top left: stacked bar, top right: table form, bottom: box plot combined with a table. Five questions about how easy the charts are to read where asked for each chart. Box plot was also asked about individually. A trainer and 18 players from his team answered the survey. The values in Table 4.2 are the mean rounded to two decimals from what the players answered. One means they thought Question X was very easy to answer and five is for very hard or impossible. If ten players would have answered a ’1’ on Question 1 for Box plot and eight players answered a ’2’, the mean is calculated as shown in equation (4.1) and.

(33) 4.2. Survey. 21. put in the corresponding table cell. The questions are referred to same numbers as in section 3.2 10 ∗ 1 + 8 ∗ 2 = 1, 44.. 18 Question Question 1 Question 2 Question 3 Question 4 Question 5 Mean Total. Box plot 3,27 1,44 1,72 2,06 2,89 2,28. Stacked barchart 2,33 2,16 1,77 2,39 3,06 2.34. Table 1,28 1,78 2,00 2,11 1,17 1,67. (4.1) Box plot + Table 2,50 1,72 2,00 1,72 1,94 1,97. Table 4.2: Mean values for each question, collected from the result of the survey done by players.. Question Question 1 Question 2 Question 3 Question 4 Question 5 Mean Total. Box plot 5 2 3 2 5 3,4. Stacked barchart 5 5 3 1 5 3,8. Table 1 1 1 1 1 1. Box plot + Table 1 1 1 1 1 1. Table 4.3: Answers for each question, collected from the result of the survey done by the trainer..

(34)

(35) Chapter 5. Analysis and Discussion. After trying a different combination of charts for the categories effort and freshness, and receiving feedback from our supervisor and the trainer, we concluded that having all needed variables present in one chart is not an acceptable solution. To represent all data variables in one single chart is too much information for the user to comprehend at a glance. Therefore the solution was to divide the variables into multiple steps of different views. After each step, the reader narrows down deeper into a single player’s data. To satisfy the need of getting an overview of the team status that is easy to understand, was box plot selected. The reason behind it was that the chart’s layout does not change if there are few data points contra multiple data points. A bubble chart would, for example, get huge bubbles that overflow into other bubbles if there were a high number of players. Scaling down the bubbles is not an acceptable solution due to that a significant gap in the player amount would lead to the smaller bubbles being downscaled to a point where they are not visible anymore. Box plot remains intact despite variances in size of the datasets, and it will still be able to show how the players’ effort are spread. It will also show where the mean and outliers are. With the help of the greyed area of expected effort, the primary view makes clear how the team relates to the trainer’s expectations. Since expected effort can vary from day to day line charts are suitable. Even if box plot is great at showing how the players are relating to each other in effort, it will not show specific distribution and to whom the data points belong. That is why a secondary view was created with the role of "zooming in" on a selected day in the box plot. For the secondary view, a table was chosen to represent the data. A table is good at showing detailed data, and therefore it contemplates the box plot’s weaknesses as excellent. When the primary and secondary view is put side by side with each other, it becomes a powerful combination of the chart types’ strengths. Since players who are within the expected effort area are not the most interesting to view, a decision was made to hide those players in the table by default. When the user has come to the tertiary view she has already narrowed down the data to a single player and box plot is no longer a valid option since it needs more than one player to be rewarding. Line chart was not a valid option for representing multiple players since the associated lines would make it too cluttered, as shown in fig 5.1, but for a single player line chart is excellent in how it shows change over time. Since effort and freshness are the attributes that the trainer is most interested in, the first chart on the quaternary view is a line chart populated with mentioned attributes and upper/lower markers for expected effort. Request from the trainer 23.

(36) 24. Chapter 5. Analysis and Discussion. Figure 5.1: An unreadable example of line chart representing 15 players’ reported effort. The reported effort is for one week. The chart was made during implementation of the web application. was to also include other training, injuries, and comments in the first charts. In that way, it was easier to see the date the entries originated of. The remaining training attributes do not need as much focus and will therefore not need to be easily attainable. Introducing them in the quaternary view will not draw attention from the user until they are wanted. The table got the lowest score in the player survey which means it was the easiest to read data from considering the questions asked. This table result must not be mistaken for the table form the trainer previously had been using which was the unprocessed raw format. Tables can be very informative but only when excluding unnecessary data and when the dataset is small or sorted. Surprisingly was box plot combined with table harder to understand than the table alone. It is likely because box plot is not commonly used and looks foreign to most people. In the combination, it might confuse more than it helps if the reader is not used to look at box plots. As long as the user of the website is familiar with box plots the combination of box plot and table should be better than the separative alternatives. There was a more scattered answer for the box plot and stacked bar chart which supports the thought that they are harder to understand for an untrained eye. Some questions in the survey are not even possible to answer for some charts but still have some low-value answers. Either the players have not understood the question/chart or have not responded truthfully. The trainer’s answers are more like how we think about the charts. One noteworthy difference is that he thought the table was very easy to read in all cases, which he also answered for the combined chart. It could mean that the combined version is redundant and unnecessary. Not to forget is that the table in the survey does not have considerably many rows and columns and is not that hard to read. If the table consisted of data from 30-40 players instead of 15, it would have been much harder to read. Visualization also adds aesthetic value and catches the eye better than a table does..

(37) Chapter 6. Conclusions and Future Work. Conclusions Of all the charts that we have tried visualizing with the dataset received from the trainer, box plot has proven to be the most suitable for effort and freshness. Box plot can give the trainer a quick and understandable overview for all his players, which the trainer was unable to do before using his method (reading from a spreadsheet). Remaining attributes were visualized for a single player with bar and line charts. When representing all variables from a dataset in a single chart are not achievable, dividing them into multiple charts that are chained together could be one acceptable solution. Adding the possibility to interact with the chart, the user can choose to look at a more detailed view if wanted. Compared to reading the data in the raw table form, processing, sorting and visualizing the data has proven to ease the understanding and analysis of given data. Tables are still useful, but the reader needs to know what she is looking for, for them to be viable. Future work Correlations between data types can be experimented with and visualized to improve the analysis further. With the help of a correlation values could harmful behavior be detected at an early stage and automated warnings be sent out to the trainer and affected player. Extending the collection of training data with more attributes such as heartbeats per minute, GPS data, and video material could provide a more comprehensive analysis. The trainer has had some problems with motivating the players to report their training data. Simplifying or even automate parts of the reporting process with cloud syncing with external applications might improve the reporting frequency. External applications could be used, for example, the software used for collecting data from activity bracelets. A study could be performed to conclude if data gathered and analyzed from training sessions help the trainer plan the sessions. Also, if it helps the players to evolve and perform better on game sessions, compared to not collect and analyze.. 25.

(38)

(39) References [1] Borg scale. http://exercise.trekeducation.org/assessment/borg-scale-rpe/. [2] Google form. https://www.google.com/forms/about. Accessed: 2018-05-10. [3] Minskade antalet anmälda skador i damallsvenskan. http://www.efd.se/ minskade-antalet-anmalda-skador-i-damallsvenskan/. Accessed: 201805-10. [4] Plotly js. https://plot.ly/javascript/. Accessed: 2018-05-10. [5] Trimbite. http://www.trimbite.com/. Accessed: 2018-05-10. [6] A. Abela. Advanced Presentations by Design : Creating Communication That Drives Action. Center for Creative Leadership, 2nd edition, 2013. [7] L. Nelsen et al C. Carling, J. Bloomfield. The role of motion analysis in elite soccer: contemporary performance measurement techniques and work rate data. Sports Medicine, Vol 38, Issue 10:839–862, 2008. [8] S. Landin C. Hedlund. Gps som tekniskt hjälpmedel inom fotbollen. Master’s thesis, Högskolan Dalarna, 2005. [9] FNF Coaches. https://fnfcoaches.com/best-apps-for-football-coaches/. https: //fnfcoaches.com/best-apps-for-football-coaches/. Accessed: 2018-0510. [10] M. Cox. Top 6 effective football coaching apps. https: //www.thesoccerstore.co.uk/blog/football-coaching/ top-6-effective-football-coaching-apps/. Accessed: 2018-05-10. [11] A. Few. The effective visual communication of data. Information dashboard design, Vol 7, Issue 2:12, 2006. [12] A. Pio D. Tore G. Raiola G. Altavilla, L. Riela. The physical effort required from professional football players in different playing positions. Journal of Physical Education and Sports, Vol 17, No 3, 2017. [13] M. Caldara M. McBride. The efficacy of tables versus graphs in disrupting dark networks: An experimental study. Social Networks, Vol 35, Issue 3:406–422, 2013. [14] R. L. Nuzzo. The box plots alternative for visualizing quantitative data. PM & R, Vol 8, Issue 3:268–272, 2016. 27.

(40) 28. References. [15] D. Andersson P. A. Albinsson. Extending the attribute explorer to support professional team-sport analysis. Information Visualization, Vol 7, Issue 2:163– 169, 2008. [16] C. Button S Barris. A review of vision-based motion analysis in sport. Sports Medicine, Vol 38, Issue 12:1025–1043, 2008. [17] S. J. Kistler T. Azzam, A. A. Germuth. Data visualization and evaluation. New Directions for Evaluation, Volume 2013 Issue 139:7–32, 2013..

(41) Appendix A. Supplemental Information. Survey. 29.

(42)

(43) .

(44)

(45)

(46)

(47)

(48) !.

(49)

(50)

(51)

(52) .

(53)

(54)

(55)

(56) . .

(57) !

(58) " ! !# $

(59) $

(60) % &

(61)

(62)

(63) '

(64)

(65) ()

(66) * *

(67) +,-.

(68) . . #/

(69) 0 00 0 .

(70) ". # $%.

(71)

(72) .

(73)

(74)

(75)

(76)

(77) !. )/1

(78) . 2 2 34! 5* 677819

(79) :;<=>

(80) ?$5. !

(81)

(82) !. ). %. . . * . * . )

(83)

(84) !!

(85) !. ). %. . . * . * . %

(86) !

(87) !. ). %. . . * . * . "

(88)

(89) ! * . &

(90) ". ). %. . * . # $%.

(91)

(92) .

(93)

(94)

(95)

(96)

(97) !.

(98) #

(99) !. ). %. . . * . * . $ &

(100)

(101) . "

(102)

(103) (=!.

(104)

(105)

(106)

(107)

(108) @ = $ . 2 2 34!)AB8C$D& %E!*A#?F BGC,;#A*. B

(109)

(110) ! * . '

(111) ". ). %. . * . # $%.

(112)

(113) .

(114)

(115)

(116)

(117)

(118) !. '

(119)

(120) !!

(121) !. ). %. . . * . * . :

(122) !

(123) !. ). %. . . * . * . A "

(124)

(125) !. ). %. . . * . * . !#

(126) #

(127) ! * . ). %. . * . %! H . 2 2 34!<&:G#'I> 'F?$ C6C:?7. (

(128) ". # $%.

(129)

(130) .

(131)

(132)

(133)

(134)

(135) !. !!

(136)

(137) !. ). %. . . * . * . !)

(138)

(139) !!

(140) ! * .

(141) ". ). %. . * . # $%.

(142)

(143) .

(144)

(145)

(146)

(147)

(148) !. !%

(149) . !

(150) !. ). %. . . * . * . ! "

(151)

(152) !. ). %. . . * . * . !

(153) #

(154) !. ). %. . . * . * .

(155) ! ! F

(156) &

(157)

(158) . 2 2 34!FCEFAIJ:.=K-8<C. )

(159) ". # $%.

(160)

(161) .

(162)

(163)

(164)

(165)

(166) !. !B

(167)

(168) !. ). %. . . * . * . !'

(169)

(170) !!

(171) !. ). %. . . * . * . !:

(172) . !

(173) !. ). %. . . * . * . !A "

(174)

(175) !. ). %. . . * . * . )#

(176) #

(177) ! * . ). %. . * . 5D. "

(178) ". # $%.

(179) 37 Result from survey. Surveyee Player 1 Player 2 Player 3 Player 4 Player 5 Player 6 Player 7 Player 8 Player 9 Player 10 Player 11 Player 12 Player 13 Player 14 Player 15 Player 16 Player 17 Player 18 Trainer. Q1 2 5 3 5 2 1 5 4 1 1 5 4 3 5 3 3 3 4 5. Q2 2 1 3 1 3 1 1 1 1 1 1 2 1 1 1 2 1 2 2. Q3 2 1 1 1 4 1 4 1 1 1 1 3 1 1 1 3 2 2 3. Q4 3 4 1 3 5 1 3 2 4 1 1 1 1 1 1 1 2 2 2. Q5 3 3 2 2 3 1 2 4 1 5 3 3 2 5 3 2 3 5 5. Table A.1: Survey result for box plot.

(180) 38. Appendix A. Supplemental Information Surveyee Player 1 Player 2 Player 3 Player 4 Player 5 Player 6 Player 7 Player 8 Player 9 Player 10 Player 11 Player 12 Player 13 Player 14 Player 15 Player 16 Player 17 Player 18 Trainer. Q1 5 3 1 1 1 1 2 1 1 1 5 5 3 4 2 2 2 2 5. Q2 5 5 1 1 1 1 4 2 1 1 3 2 1 5 1 2 1 2 5. Q3 3 3 1 1 2 1 4 1 1 1 1 2 2 1 2 2 2 2 3. Q4 4 3 1 3 3 1 3 4 3 1 3 1 1 1 3 3 2 3 1. Q5 3 3 2 4 3 1 2 3 1 1 5 4 4 5 2 3 5 4 5. Table A.2: Survey result for stacked barchart. Surveyee Player 1 Player 2 Player 3 Player 4 Player 5 Player 6 Player 7 Player 8 Player 9 Player 10 Player 11 Player 12 Player 13 Player 14 Player 15 Player 16 Player 17 Player 18 Trainer. Q1 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 3 2 1. Q2 1 2 2 1 2 4 1 2 1 1 1 4 1 1 2 1 3 2 1. Q3 2 2 3 3 2 4 3 3 1 1 1 2 1 1 1 2 2 2 1. Q4 1 2 2 3 3 4 4 2 1 1 1 4 1 1 2 3 2 1 1. Q5 1 1 1 2 1 1 1 1 1 1 1 2 1 1 1 1 2 1 1. Table A.3: Survey result for table.

(181) 39 Surveyee Player 1 Player 2 Player 3 Player 4 Player 5 Player 6 Player 7 Player 8 Player 9 Player 10 Player 11 Player 12 Player 13 Player 14 Player 15 Player 16 Player 17 Player 18 Trainer. Q1 3 2 3 1 3 1 4 4 2 1 1 5 1 5 2 2 2 3 1. Q2 3 1 2 1 3 1 3 2 1 1 1 2 1 1 1 3 2 2 1. Q3 3 2 2 1 2 1 4 3 1 3 1 2 1 1 2 3 2 2 1. Q4 3 2 1 1 3 1 3 2 1 1 1 2 1 1 1 3 2 2 1. Q5 1 2 2 1 1 2 3 4 1 1 1 2 1 5 1 2 2 3 1. Table A.4: Survey result for combination of box plot and table Dataset used to generate charts for survey {" t r a i n e r " : " Herman " , " team " : " K a r l s k r o n a " , " p l a y e r s " : [ {"name " : " S p e l a r e A" , " data " : [ { " da t e ":"2018 −03 −27" ," e f f o r t " : { " l e v e l " : " 7 " , " d i f f " : 3 } } , { " dat e ":"2018 −03 −30" ," e f f o r t " : { " l e v e l ":"4" ," d i f f ": −1}}]} , {"name " : " S p e l a r e B" , " data " : [ { " dat e ":"2018 −03 −26" ," e f f o r t " : { " l e v e l " : " 4 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −27" ," e f f o r t " : { " l e v e l " : " 4 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −29" ," e f f o r t " : { " l e v e l " : " 7 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −30" ," e f f o r t " : { " l e v e l ":"7" ," d i f f ":0}}]} , {"name " : " S p e l a r e C" , " data " : [ { " dat e ":"2018 −03 −26" ," e f f o r t " : { " l e v e l " : " 5 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −27" ," e f f o r t " : { " l e v e l " : " 4 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −30" ," e f f o r t " : { " l e v e l ":"6" ," d i f f ":0}}]} , {"name " : " S p e l a r e D" , " data " : [ { " da t e ":"2018 −03 −26" ," e f f o r t " : { " l e v e l " : " 5 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −29" ," e f f o r t " : { " l e v e l " : " 8 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −30" ," e f f o r t " : { " l e v e l ":"8" ," d i f f ":1}}]} , {"name " : " S p e l a r e E" , " data " : [ { " dat e ":"2018 −03 −26" ," e f f o r t " : { " l e v e l " : " 5 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −27" ," e f f o r t " : { " l e v e l " : " 5 " , " d i f f " : 1 } } , { " dat e ":"2018 −03 −29" ," e f f o r t " : { " l e v e l " : " 8 " , " d i f f " : 0 } } , { " dat e ":"2018 −03 −30" ," e f f o r t " : { " l e v e l ":"7" ," d i f f ":0}}]} ,.

No results found