Data-Driven Prediction of Track Quality Index (TQI): A Comparative Study of Statistical and Ensemble Learning Models – A Case Study of the Kolašin–Podgorica Railway

Maintaining the geometric quality of railway tracks is essential for ensuring the safety and efficiency of railway operations. This study presents a comparative analysis of Multiple Linear Regression (MLR) and Random Forest (RF) models for predicting the Track Quality Index (TQI), based on historical inspection data collected from the mountainous Kolašin–Podgorica railway section between 2017 and 2022, with data from 2024 reserved for independent validation. The dataset includes high-resolution measurements divided into 20-meter homogeneous units, incorporating infrastructure, geometric, operational, and maintenance-related variables. Both models were trained on scaled input features, and their performance was evaluated using the coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE). The results demonstrate that the machine learning approach significantly outperformed the statistical baseline; the RF model achieved a higher goodness-of-fit (R2 = 0.69 vs. 0.57) and reduced the average prediction error (MAE) by approximately 15% compared to the MLR model. Furthermore, RF exhibited superior stability in capturing severe localized degradation trends. These findings highlight the potential of ensemble machine learning methods to mitigate large prediction errors and enhance data-driven, proactive track maintenance planning in geometrically complex railway networks.