Optimizing endless runner game player performance using a hybrid GMF-MLP recommendation system based on neural collaborative filtering

Gunawan, Hendry Cahyo, Nugroho, Fresy ORCID: https://orcid.org/0000-0001-9448-316X, Yaqin, Muhammad Ainul, Suhartono, Suhartono and Arif, Yunifa Miftakhul ORCID: https://orcid.org/0000-0002-2183-0762 (2026) Optimizing endless runner game player performance using a hybrid GMF-MLP recommendation system based on neural collaborative filtering. Jurnal Teknologi dan Manajemen Informatika, 12 (1). pp. 1-9. ISSN 2580-8044

Preview

Text 29376.pdf - Published Version Available under License Creative Commons Attribution Share Alike. Download (963kB) | Preview

Abstract

Background : Endless Runner genre games have a game difficulty level that increases exponentially as the distance traveled increases, causing character failure and player frustration. The main problem lies in the inability of players to choose supporting items that are appropriate to the context of difficulty. Previous research in recommendation systems generally focuses on purchase prediction (monetization). Objective : This study aims to build a personalized item recommendation system to minimize failure and maximize scores. This study proposes a hybrid architecture of General Matrix Factorization (GMF) and Multi-Layer Perceptron (MLP) within the Neural Collaborative Filtering (NCF) framework to model linear and non-linear interactions between player playstyle and failure context (death cause). Method : Using the NCF approach with a hybrid architecture : General Matrix Factorization (GMF) to capture linear preferences and Multi-Layer Perceptron (MLP) to model non-linear interactions of game features. The dataset consists of 10,000 game activity logs that include variables such as jump count, obstacle avoided and death cause. Results : Evaluation of the case study shows the performance of the machine learning model during the training process of 20 epochs with both lines (Train & Val) ending at around 0.88 - 0.90 (88% - 90%). The distance between the blue line (Train Accuracy) and the orange line (Val Accuracy) is very small. This is an indicator that the model does not experience significant overfitting. Conclusion : The integration of GMF and MLP has proven effective in providing recommendations that are adaptive to gameplay dynamics.

Item Type:	Journal Article
Keywords:	endless runner; general matrix factorization ; multi-layer perceptron; neural collaborative filtering; game recommendation system
Subjects:	08 INFORMATION AND COMPUTING SCIENCES > 0802 Computation Theory and Mathematics > 080299 Computation Theory and Mathematics not elsewhere classified 08 INFORMATION AND COMPUTING SCIENCES > 0899 Other Information and Computing Sciences > 089999 Information and Computing Sciences not elsewhere classified
Divisions:	Biro Administrasi Akademik, dan Kemahasiswaan
Depositing User:	Hendry Cahyo Gunawan
Date Deposited:	28 Oct 2026 10:22

Downloads

Origin of downloads

Actions (login required)

View Item