Dukungan Keputusan dalam Permainan Berbasis Giliran untuk Panduan Pertempuran dan Pembentukan Partai Dengan Menggunakan Ai dan Deep Reinforcement Learning

Abstract

Granblue Fantasy is a turn-based game that demands complex strategy. However, the pay-to-win system often hinders the progress of free-to-play players with limited resources. This research develops a recommendation system based on Deep Learning with Hybrid Dyna-Q Inspired Deep Reinforcement Learning (DRL) method for selecting the best actions during battles, as well as Deep Learning with Supervised Deep Learning Classification method for optimizing team composition. By utilizing the Markov Decision Process (MDP), the system can simulate various battle scenarios and generate more accurate recommendations. Experimental results show that the hybrid approach combining DRL and MDP improves strategic effectiveness and character selection compared to conventional methods. Action Suggestion received an average score of 5/5, demonstrating accurate predictions for skill usage timing and conditions. Party Suggestion scored 5/5 for the clarity of character selection but only 3/5 for composition alignment with player strategies and 2/5 for adoption in actual gameplay, particularly among higher-ranked players. The system is not designed to replace player decision-making but serves as an assistive tool for those struggling to optimize their limited resources. This research highlights the potential of DRL in complex and dynamic decision-support systems.