Model Reuse With Bike Rental Station

4834 Words20 Pages
Model Reuse with Bike Rental Station Data Authors: 1. Arun Bala Subramaniyan, M.S. Industrial Engineering, Arizona State University. 2. Dr. Rong Pan, Associate Professor of Industrial Engineering, Arizona State University. Introduction and Motivation Bike Rental Stations are a good business in places with large number of tourists and also the native people rent bikes for their day to day work. In this project, the bike rental station located in Valencia, the third largest city of Spain is considered. The bike rental company would like to predict the number of bikes available in each station three hours in advance. There are atleast two uses for such prediction. At first, a user plans to rent (or return) a bike in 3 hours time and wants to…show more content…
This process is continued for selecting the best model for all the new stations (201 to 275). The R software package is used for this purpose. The extracted best models for the new stations are stored in .csv file. In some cases, the prediction result is negative or it exceeds the maximum limit of the bikes that can be parked in a station. This can be overcome by adding a constraint such that whenever the result is negative, the value is reset to zero and whenever the result exceeds the maximum limit, the value is reset to the number of docks at the particular station. So, this helps in reducing the error value. Prediction Using the extracted models, the number of bikes at the new stations is predicted. The same constraints are applied to avoid negative values and over fitting. The R software is used for predicting the number of bikes. The results of this prediction are stored in .csv file. Other Methods tried for prediction Instead of reusing the trained models, new models were built with the given deployment data for stations 201 to 275. Some of the methods used are given below. Ordinary Least Squares Method After collecting and cleaning the data, the first model was built using all the regressors under consideration. A thorough analysis of this full model, including residual analysis and multicollinearity check was done. The best subset regression was also tried. The normal probability
Open Document