Forecasting passenger flow volumes on Serbian railways: A comparative analysis of univariate and multivariate approaches

Railway passenger transport in Serbia has seen a significant increase in recent years, driven by substantial investments in infrastructure and rolling stock. Under these evolving conditions, improving operational efficiency, planning, and resource allocation has become essential. Accurate passenger volume forecasting serves as a critical tool for proactive decision-making, enabling railway operators to remain flexible and effectively respond to fluctuating demand. A range of univariate and multivariate forecasting methods was applied to generate medium-term predictions of passenger volumes for the Serbian railways. Traditional univariate linear Seasonal Autoregressive Integrated Moving Average (SARIMA) model was compared with both univariate and multivariate forms of the Long Short-Term Memory (LSTM) neural network, SARIMA-LSTM model and Genetic Algorithm-optimized GA-LSTM model. A comparative analysis using various performance measures demonstrates that multivariate methods outperform their univariate counterparts. Among them, the multivariate GA-LSTM model achieves the highest predictive accuracy, making it the most effective approach for addressing challenges related to planning inefficiencies and suboptimal resource utilization.