Multi-step-ahead forecasting of bike-sharing demand using multilayer perceptron model with additional timestamp features

Abstract

Bike sharing is increasingly gaining popularity as an affordable and environmentally friendly mode of transportation in urban areas. However, the nature of bike sharing, where users can pick up and return bikes at different stations, often results in an uneven distribution of bikes across stations. Consequently, accurately predicting the future number of rented bikes at each station becomes crucial for bike-sharing operators to optimize the bike inventory at each location. This study introduces a multi-step-ahead forecasting model that employs machine learning methods to predict the hourly demand for rented bikes. We utilize information on rented bikes from the preceding day to forecast the forthcoming counts of rented bikes for the next 1, 3, 6, 12, and 24 h. Additional features extracted from timestamps are incorporated to enhance the accuracy of the model. We compare the proposed model, based on multilayer perceptron (MLP), with various machine learning prediction algorithms, including Support Vector Regression (SVR), K-Nearest Neighbor (KNN), Decision Tree (DT), Adaptive Boosting (AdaBoost), Random Forest (RF), and Linear Regression (LR). Applying the proposed MLP model to the Seoul bike-sharing dataset demonstrates a positive outcome, indicating a reduction in prediction error compared to other forecasting models. The proposed model achieves the highest R-2 (coefficient of determination) values when compared to other models, with values of 0.973, 0.882, 0.82, 0.807, and 0.79 for prediction horizons of 1, 3, 6, 12, and 24 h, respectively. By obtaining future values for predicted rented bikes, the trained model is anticipated to assist in optimizing the number of available bikes for bike-sharing companies.