Forecasting Licensed Athletes Numbers Using Time Series and Hybrid Artificial Intelligence Models

Abstract

This study comparatively applied four distinct time series and artificial intelligence-based forecasting models to predict the short-term number of licensed athletes in Türkiye: Autoregressive Integrated Moving Average (ARIMA), Convolutional Neural Network with Long Short-Term Memory (CNN-LSTM), Extremely Boosted Gradient Decision Trees with Long Short-Term Memory (XGBoostLSTM), and gated recurrent unit with random forest (RF-GRU). Models were trained using data from 2005 to 2019; forecasts for the years 2020-2024 were subsequently compared with actual values. The ARIMA model demonstrated the highest predictive efficacy, achieving a coefficient of determination of 0.9633, a mean square error of 0.1056, and a mean absolute error of 0.1990. The CNN-LSTM model exhibited a coefficient of determination of 0.9771 and a mean absolute percentage error of 5.68%. The remaining two hybrid models (XGBoost-LSTM and RF-GRU) exhibited inferior accuracy, accompanied by comparatively elevated error levels. The results underscore the significance of data-driven decision-making in sports policy and provide scholarly contributions to strategic planning.

Received: 28 September 2025 | Revised: 25 November 2025 | Accepted: 11 December 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work

Data Availability Statement

Data are available from the corresponding author upon reasonable request.

Author Contribution Statement

Halil S¸enol: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Writing - review & editing, Visualization, Supervision, Project administration. Halil Çolak: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Writing - review & editing, Visualization, Supervision, Project administration.