Forecasting Daily Ridership for KTMB Services in Malaysia By Using Time Series Analysis

Abstract

Proper prediction of the ridership in public transport is critical in operational planning, capacity management as well as optimisation of service. Daily ridership change, which is affected by weekends and weekdays and seasonal and special events, is one of the issues that demand prediction becomes a daunting task to rail operators. Poor forecasts can cause resource wastage, excess capacity or under-utilization of the services. Thus, this study compare, evaluate and analyse the time series forecasting models used to predict daily KTM Komuter ridership in Malaysia. The objectives of this study are to analyse the time series forecasting models Naïve, SARIMA, and Prophet using daily KTMB ridership data, to evaluate and compare the performance of these models using accuracy metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), and to forecast future daily ridership using the best-performing model. Three forecasting approaches were considered: the Naïve method, Seasonal Autoregressive Integrated Moving Average (SARIMA) and Facebook Prophet. First, the exploratory data analysis was performed to define the main attributes of the time series that is trend, variability, and the weekly seasonality. In the case of the SARIMA model, non-seasonal and seasonal differencing was used to obtain stationarity and suitable model structures were found based on ACF and PACF plots and then diagnostic checking was done. Facebook Prophet model was used to extract the trend and seasonal variations whereas Naïve method was used as a control baseline. The findings show that the SARIMA  model had the smallest errors in the test phase, proving it to be more precise and consistent than the Naïve and Facebook Prophet models. During the testing stage, the SARIMA model obtained the minimum percentage error in MAPE of 8.94%, which was lower than Facebook Prophet with a significantly high percentage error at 93.99%. Even though Facebook Prophet handled trends and seasonality well, it struggled with predictions beyond the training data. Ultimately, the results suggest SARIMA is the top pick for predicting daily ridership, delivering dependable forecasts to aid in operations and strategy.