Predicting Public Bicycle Rental Number using Multi-source Data

In order to solve the problems about "no bicycle to rent" and "no place to return", Public bicycle system (PBS) requires rental amount prediction. We build a hybrid short-term rental prediction model of public bicycle by using public bicycle renting record data and environmental characteristic data such as holiday, weather and temperature. By K-means cluster analysis, different rental modes of public bicycle are distinguished. we extract the environmental characteristic rules to build the Bayesian classifier by using the datasets. The data samples of different rental modes are used to training the adaptive particle swarm optimization-back-propagation neural network (APSO-BP). When predict public bicycle rental amount, choose the APSO-BP model with the same rental mode to predict the rental amount of public bicycle during the next time period. We use the real Hangzhou public bicycle rental data to verify the validity of the method. The average prediction accuracy is over 95%.