Analytic Hierarchy Process Weighted Logistic Regression Model for League of Legends – Kamal (ERNC25) скачать в хорошем качестве

Analytic Hierarchy Process Weighted Logistic Regression Model for League of Legends – Kamal (ERNC25)

This presentation by Ahmad Alif Kamal from Universiti Sains Malaysia examines how expert-informed decision frameworks can strengthen predictive analytics in competitive esports, using League of Legends as a data-rich case study. Responding to the limitations of conventional machine learning models in capturing the strategic complexity of esports, the study integrates the Analytic Hierarchy Process (AHP) with multiple logistic regression (LR) variants. AHP was used to weight key early-game performance dimensions—combat effectiveness, farming efficiency, objective control, survivability, and vision—based on domain expertise, transforming multi-dimensional in-game statistics into a single interpretable aggregate feature. Using a publicly available dataset of high-diamond ranked League of Legends matches, the study evaluated eight logistic regression model variants, including baseline, regularized, hierarchical, multinomial, and ensemble approaches. Model performance was assessed using standard classification metrics across multiple train-test splits, alongside ablation analyses to measure the added value of the AHP-derived feature. Findings show that ensemble, hierarchical, and multinomial logistic regression models achieved the strongest predictive performance. Incorporating the AHP-weighted feature consistently improved results, with objective control and vision emerging as particularly influential factors in early-game win prediction. The findings highlight the value of combining statistical modelling with expert knowledge in esports analytics. Overall, this research demonstrates how hybrid, interpretable modelling approaches can bridge the gap between data science and in-game strategic understanding. It provides practical insights for analysts and coaches while contributing to the development of explainable and domain-sensitive analytics in esports. Keywords: Analytic Hierarchy Process, machine learning, League of Legends, logistic regression, predictive modelling About the ERN Conference 2025: The Esports Research Network Conference 2025 took place online from November 12–14, 2025 as a continuous, 72-hour global livestream. Guided by the theme “Future Realities: Esports as a Global Lab,” the event explored how esports serves as a living laboratory for understanding digital transformation across education, media, technology, and society. This innovative digital-first format followed the sun across three regional hubs, Slippery Rock University of Pennsylvania & Michigan State University (Americas), Staffordshire University (EMEA), and UNSW Sydney (Asia-Pacific), creating a nonstop, interconnected academic experience available to participants worldwide. This ambitious production was made possible through the support of Corsair for Business, whose Elgato technology powered high-quality broadcasting and seamless livestreaming across all three hubs. ERNC25 continues the network’s mission to unite scholars, practitioners, and students while fostering open, global collaboration in esports research. Full abstracts for all presentations can be viewed in the official Book of Abstracts at: https://book.esportsresearch.net Website: https://esportsresearch.net