Weather Forecasting Based Shared Bike Demand Analysis using Machine Learning

Abstract

This paper describes the status of US bike-sharing provider named BoomBikes who has recently suffered considerable dips in their revenues because of the pandemic. The company is not able to sustain itself. Hence, This paper analyze the application of various machine learning techniques like Linear Regression, Polynomial Regression, Lasso Regression, Ridge Regression, XGBoost Regressor, K-Nearest Neighbor, Decision Tree, Gradient Boost Regressor, Random Forest, ADA Boost Regressor and Support Vector Machine for predicting the future sales of the BoomBikes. The work is divided into parts i.e. first by considering all the features provided and second by reducing the dimensions using Principal Component Analysis. For better analysis and understanding, we have reduced the number of columns in dataset from 16 to 8. After analyzing all the Regression models, we found that Linear Regression and Polynomial Regression have best R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.98 in case of first approach while the value of Linear Regression drops more after applying Principal Component Analysis compared to Polynomial Regression with values 0.8844 and 0.9065 respectively. Taking both the approaches into consideration, we have selected Polynomial Regression as the best cited model. Talking about the worst model, K-Nearest Neighbor has the least R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.9334 in case of first approach and Decision Tree with R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.8227 in the second approach. Also, the Mean Squared Error, Mean Absolute Error and Root Mean Squared Error are also calculated for the above used algorithms.