This study represents an effort to enhance player engagement by dynamically balancing game difficulty using transfer learning and Long Short-Term Memory(LSTM) networks. Traditional difficulty adjustment methods are often static and manually configured, leading to decreased player engagement, higher dropout rates, and reduced cognitive learning. The proposed system personalizes gameplay by analyzing player performance metrics and adjusting real-time difficulty parameters to create a more adaptive experience. Player performance data from the third-person shooter game PinGun, including shot counts, event classifications, and timestamps, was collected and used to train an LSTM model. Transfer learning enables the model to leverage previously acquired behavioral patterns, adapting dynamically to new players. Difficulty is adjusted by modifying gameplay parameters such as enemy spawn rates, rest periods between waves, and enemy reaction speeds based on individual player performance trends. An experiment involving 15 players with varying skill levels demonstrated the effectiveness of this approach. Novice players improved their shooting accuracy by 23%, reduced reaction times to threats by 18%, and increased their level completion rates by 17%. Intermediate players exhibited a 15% improvement in task-switching efficiency and a 14% increase in success rates during challenging waves. Expert players maintained high levels of engagement, with a 5% improvement in reaction speeds, ensuring that the challenge remained appropriate without negatively impacting their overall experience.
Kheyroddin et al. (Fri,) studied this question.