Explainable AI for Analyzing Behavioral Patterns in Serious Games: From Educational to Professional Training Contexts

Abstract

This paper explores how Explainable AI (XAI) techniques, applied to machine learning models like Transformers and LSTMs, can help analyze behavioral patterns in gaming. Drawing on insights from the field of educational games for neurodivergent children - where explainability is crucial for success - We propose adapting these approaches to serious games for professional training. By capturing worker engagement and explaining performance outcomes, XAI can support not only the workers themselves but also instructors and managers in comprehending why failures occur. Our experiments with Portuguese gaming datasets show that LSTM models, combined with explainability tools like LIME, outperform Transformers with a relative gain of 7% for LSTM, when identifying nuanced emotional patterns. The obtained metrics from the validation phase are an accuracy of 0.78 for the LSTM model and 0.73 for the Transformer. We argue that such explainable insights can enhance professional training by providing more targeted feedback and improving the overall training process.

Ähnliche Arbeiten