| Summary: | Train delays can have a significant impact on transport systems, causing passengers inconvenience and operational challenges for train operators. One of the causes of railway delays is seasonality, which includes holidays and weather patterns. ARIMA model is a well-known statistical model for analyzing and forecasting time series data. However, the ARIMA model is unable to deal with seasonal time series data. Furthermore, time series data frequently exhibit stochastic patterns that may include outliers, whereas the traditional ARIMA model fails to produce accurate results when the data contains outliers. This study aims to forecast train delays based on Network Rail dataset using Facebook (FB) Prophet. FB prophet is a flexible time series model that is renowned for its ability to deal with seasonality and outliers. The number of 713 daily delayed occurrences (in minutes) from Edinburgh, Scotland was taken into account and divided at a ratio of 70:30 for model fitting and evaluation, respectively. The testing dataset was fitted with ARIMA, FB Prophet with and without holidays effect. The mean absolute error (MAE) and root mean squared error (RMSE) were used to evaluate the models. The RMSE and MAE of the FB prophet model with holidays feature deliver best performance in term of forecasting accuracy with the lowest rate at 968.446 and 638.558 respectively, as compared to ARIMA and FB prophet without holidays feature. The results are further improved after removing the outliers. This study highlights the potential of FB Prophet model with holidays feature as a predictive analytic tool to enhance train services. © 2024 Author(s).
|
|---|