Fare revenue forecast in public transport: A comparative case study

This paper presents results from a case study of fare revenue prediction in public transportation in Berlin using machine learning and time series analysis. Our work aims to aid in the implementation of automated revenue controlling and data-driven decision support within existing controlling processes.

We generate forecasts based on fare revenue data for different product segments aggregated on a monthly basis. Additionally, we model exogenous effects using data publicly available.

The results were obtained using a variety of methods including regression methods as well as autoregressive methods and exponential smoothing. Among others, SARIMAX, MLR, LASSO and Ridge were applied.

We evaluate the predictive quality of each method and compare them. Where appropriate, we apply automatic feature selection to improve performance.

Our findings, alongside a discussion of their interpretability, can serve as recommendations for practitioners, supporting them in choosing appropriate methods and suitable exogenous variables to reliably predict the fare revenues of different products.