Interpretable Machine Learning for Intelligent Transportation in Bike-Sharing

Abstract

In recent years, the benefits of bike-sharing have become increasingly clear, with cities around the world benefiting from increased road resource utilisation, reduced traffic congestion, and improved urban mobility. Bike-sharing has proven to be an essential mode of transportation and a cornerstone of smart city initiatives. To ensure that bike-sharing service providers can provide an optimal experience to their customers, it is crucial to have accurate information about the total number of bikes available at each station, as imbalances caused by bike shortages can negatively impact the service. While many previous studies have used machine learning techniques to predict demand, few have addressed how these models make their predictions.This study focuses on developing explainable and interpretable models for predicting the total number of bikes. We employ a range of regression methods, including Linear Regression (LR), Support Vector Regression (SVR), Random Forest (RF), and Gradient Boosting Regressor (GBR), to accurately forecast the distribution of bikes in a bike-sharing service. Our approach uses Exploratory Data Analysis (EDA), model selection, validation, and Explainable Artificial Intelligence (XAI) to provide a clear and transparent interpretation of the model predictions.To evaluate the models, industry-standard metrics such as Mean Absolute Error, Mean Squared Error, and R2 Score were used. The report showed an outstanding average R2 Score of 99%, demonstrating the efficacy of the models in accurately predicting the total number of bikes in a bike-sharing service. The approach used offers a transparent and interpretable methodology for predicting bike distribution, which can aid bike-sharing service providers in providing optimal service to their customers.