Measuring Mental Fatigue for Gaming

It is commonly understood that fatigue has a detrimental effect on performance. Just like a marathoner runner struggles to sustain pace in the closing stages of a race, mental fatigue can reduce the performance of an office worker, a student, or a gamer.

At MaddCog, we help gamers understand how to improve mental performance so they can play better. When we talk to gamers, the common feedback relates to not knowing when to stop playing. They know that playing for long periods can reduce performance, but should they stop after an hour, 3 hours or more? MaddCog can help answer that question.

The Outcome

Let's jump to the end of the story first. MaddCog developed a way of monitoring mental fatigue using consumer grade sensors. EEG measures brain wave activity, while heart rate variability assesses autonomic nerve system response. Both have been shown to change with mental fatigue.

Combining this sensor data into an individualized machine learning model, allowed for mental fatigue to be measured with 82.2% accuracy.

The details are explained below, but in short, what this means is that live mental fatigue can be assessed with good accuracy using cheap and comfortable sensors. This can then be used to answer the gamers questions of when they should stop playing to avoid poor performance.

How we got there

The biggest challenge in assessing mental fatigue is that there is no gold standard to compare to. Therefore, controlled testing must be established to elicit a fatigue state. Given MaddCog’s focus on gaming, a repetitive mental challenge was used that was associated with computer game performance. This challenge was mentally intense and was repeated for 60 minutes. MaddCog had subjects undertake this one-hour test while wearing EEG sensors and a heart rate monitor.

Changes in the data throughout this test were used to establish a model which identified low fatigue and high fatigue states on a continuous scale.

Let's look at some of the more detailed results. If we assume 1 is a very fatigued state and 0 is very low fatigue, then the average fatigue level near the start of the 1-hour test was 0.33, while towards the end it was 0.76.

We can also compare predicted fatigue versus the user rated fatigue. This is shown in the graph below, which highlights the strong relationship between the two. Although, in looking at this data in depth, it also became apparent that subjects struggle to rate their fatigue with much resolution. This is the reason the simple three-point scale was used, and why it is unreliable to base decisions purely on user rated fatigue levels.

Heart rate variability (HRV) was identified as the most important variable in determining fatigue state. This isn’t surprising given HRV is a stable, slow changing metric that can provide good quality data using consumer grade sensors. Because of this, MaddCog developed a mental fatigue measure based only on HRV that is 75.4% accurate. While it is definitely more accurate with EEG, this highlights the importance of HRV as a tool in assessing mental fatigue.

The graph below shows the average predicted fatigue of all the 1-hour tests performed. As you would expect, it isn’t a linear increase. In this case, a 30 second break every 10 minutes during the test offered small recovery bouts. Additionally, it was common to see fatigue decrease towards the end of the test. A similar change was reported in published time on task data, with the explanation being that the subject knew the test was almost over and started to relax.

Game testing (the exciting bit)

The more important aspect of this article is around the assessment of mental fatigue in a gaming situation. This can be difficult to assess objectively given the free-flowing nature of a game. Therefore, we had serious gamers undertake 5+ hours of League of Legends game play, with some repeating this 5-hour session twice. During these sessions we measured a variety of mental performance metrics including mental fatigue, game performance, game durations, and rating of mental fatigue after each game.

This data showed some interesting outcomes:

• Mental fatigue increased during 86% of the games played
• Mental fatigue decreased during 83% of the breaks between games
• Mental fatigue increased 4 times as much when the player won as when they lost
• Mental fatigue was lower at the start of games that the player went onto win
• There was a weak correlation between break duration and fatigue at the start of the next game (r=0.238)

Overall, this is very insightful as it shows the mental fatigue model behaviors as expected, but also that mental fatigue can have an impact on game performance. It isn’t surprising that you become more fatigued when you win. Just like the marathon runner who increases his pace and is thus more fatigued, the winning gamer often works harder during the match and thus fatigue is often associated with a good outcome.

The other important insight is that to play well, it is advantageous to start less fatigued. Once again, this isn’t surprising. If you start a game in a more fatigued state, you hit your fatigue limit early and thus performance is impaired.

Interestingly, during the 5 hours of gaming, fatigue did no increase consistently throughout the session. Rather, it varied based on the individual games as shown in the graph below which is an example from one player. This is because the mental intensity throughout the session is not consistent. Some games are more taxing than others, and the break between each game allows the gamer to undertake substantial recovery.

Looking at this example we can see how mental fatigue changed during games and the breaks. Some interesting things of note in this example include:

• Fatigue was higher at the end of every game than the start, while it dropped in all breaks between games
• On some occasions, recovery continued into the early part of the game with fatigue continuing to decrease before game demand appears to have kicked in
• The last game is very interesting as fatigue peaked and then dropped rapidly. We see this quite often, especially when a loss seems inevitable, and thus the player starts to ease up

Conclusion

The result of this research shows that mental fatigue can be measured with suitable accuracy, it does have an impact on game performance, but it does not happen in a uniform manner. Rather, each gaming session varies and thus you cannot apply a standard rule such as just playing a set number of games to avoid fatigue.

The MaddCog approach is to understand how much your performance is impacted by mental fatigue, and then monitor fatigue levels, notifying you to stop playing when fatigue reaches a leave that performance is likely to decrease.