A way to make video games engaging is by adjusting the difficulty of the game, either making it harder or easier to increase the playability of the game. Another factor to consider is that everyone plays the same game differently and they develop their style in the game. Eventually, they get comfortable with that style of play and this project tried to give an alternate approach to see how you can detect these styles of play and change the difficulty in such a way that is specifically catered towards that play style. This would force the player to produce new ways to be able to complete the same task enabling a new experience even though the player has played the game multiple times before that. The project analyses different ways in which this can be done.
In this project, a game was built that resembled the games “Super Mario” and “The No Internet Dino” game which gave players a degree of familiarity with minimal controls and actions where the player could jump, crouch, and attack. Each of the inputs that the player performs is kept track of along with other parameters like how much power they have left, the number of lives, and the enemy pattern for the subsequent iterations. This data was then used to train a Hidden Markov Model, by using the input patterns and the enemy patterns to see if we could predict the future actions and the future enemy states that the player will have to face. A total of 8 participants took part in the experiments, out of which only 7 participants did the experiment correctly where each participant submitted the data collected over 5 iterations of play-throughs for the game which they did in their own time.
The data showed that most players preferred to use the jump action rather than the other actions to dodge the enemies moving toward them. Three distinct patterns emerged when comparing the frequency of the inputs, using these one can deduce that some players button-smash i.e. they press inputs continuously regardless of the action, the opposite where actions are performed only when needed and the balanced where the player moves when they feel like it, certain participants used the attack and crouch actions more than the rest to navigate through the enemies. The HMM for the input patterns could not predict future actions with dependability since an abundance of a NULL state caused overfitting. The enemy pattern model performed with an average prediction accuracy of 12.63%. Pearson’s correlation coefficient was also calculated between the number of lives left and the power ability usage to see if there was a correlation, two participants had a positive correlation while others had a negative correlation.
Since the data analysis shows that there is a considerable amount of variation among a small group of people it will be possible to determine player behavior on a large scale but will require the collection of data and the type of game will also play an important part in this. The use of an Artificial Intelligence model would need more in-depth knowledge of the field to do the tasks that are needed. Some factors to consider would be the different game mechanics that are integrated into the game, the abilities that the user has control over, the complexity of the enemy AI, and the different input patterns and play-styles. Understanding how these factors affect the game and how the user interprets them will help determine what and how a Dynamic Difficulty Adjustment system can be built to accommodate for the fact that each player has their way of playing the game and can be catered towards that. This will also add to the personalized experience of gameplay.
This project was conducted as a study to try and give a different take on the analysis of different play-styles from the perspective of Dynamic Difficulty Adjustment. The presence of Artificial Intelligence in different games has been implemented differently and to see if this can be done differently, if taking into consideration the different play-styles of players and the different mechanics in the game, the same level in the game will be different for different people which can be catered personally towards that specific player. If an agent is implemented to analyze the different play-styles and change the difficulty accordingly, then the player can play the game but will be playing against themselves. The type of game and the different implementations and approaches play an important role in this.