Predict Mobile Legends Match Results Based on In-Game Conditions Using Deep Learning Models

Mobile Legends, a popular Multiplayer Online Battle Arena (MOBA) game, relies on team strategy and hero combinations to influence match results. Players and teams seek to enhance their performances by analyzing past matches, understanding opponents' tendencies, and planning appropriate strategies. However, many players face challenges in maintaining their rankings due to frequent losses. This study aims to predict match outcomes based on in-game conditions to help players make better strategic decisions. The research consists of five stages: data collection, data analysis, data preprocessing, data processing, and evaluation. Data is collected from random streamers, primarily those with Indonesian flags, including match duration, number of kills, and hero selections. The preprocessing stage involves selecting and converting attributes and set roles. A deep learning model using neural networks is applied to predict match results, evaluated by accuracy, precision, recall, and an F1-score metrics. The model achieved an accuracy of 89.62%, precision of 91.68%, recall of 93.13%, and F1-score of 92.34% with the rectifier activation function providing the best performance. These results indicate that deep learning can effectively enhance prediction models to facilitate better decision-making in games.