Peak Hour Demand Prediction for Sharing Bikes: A Comparative Analysis of Performances of Machine Learning Models

Currently, many big cities have adopted the use of rental bikes to improve mobility, comfort, and quality of life by easy means of transportation. In recent years, bike-sharing programs have grown in popularity as a practical and economical means of mobility for city dwellers and visitors. The number of trips taken, the length of each ride, and the stations visited are just a few of the numerous use statistics that bike-sharing businesses gather. This information may be used to understand how people commute and to guide potential growth for bike infrastructure. It is essential that rental bike accessibility to everyone at the appropriate moment minimizes waiting time and lowers the value of carbon in surroundings. Eventually, it becomes understandable that sustaining a consistent supply of rental bikes throughout the city is a concern. It is essential to determine how many bikes will be needed for maintaining a consistent supply of rental bikes every hour. Machine learning algorithms are utilized to overcome the challenges in forecasting the demand for hourly bike rentals. To anticipate the hourly demand for bike-sharing, this study employs seven machine learning models: linear regression, Huber regression, ridge regression, extra tree regressor, decision trees, random forest, and gradient technique. The investigation made use of data on the meteorological conditions, including temperature, wind speed, visibility, dew point, humidity, snowfall, rainfall, and the number of bikes leased every hour. The dataset for this project has been collected via Kaggle. The experimental findings indicate that the extra tree regressor model, out of the seven models, produces the best R2 (0.905832). However, other models also demonstrate a considerable degree of accuracy in predicting the demand for bike-sharing.