Random Forest Prediction with Improved Feature Selection to Shared Bicycle Demand

Under the promotion of green travel, shared bicycles are developing rapidly, but urban road space is wasted due to unreasonable planning. In order to accurately place the number of bicycles, this paper predicts the demand for shared bicycles based on factors such as weather, seasonality and temperature and humidity. Faced with the complexity and collinearity of data features, a random forest prediction shared bicycle demand model with improved feature selection is proposed. First, features with collinearity are excluded by partitioning the feature saliency and correlation coefficient values. Then, the data is effectively characterized. Therefore, the upper bound of the generalization error of the algorithm is reduced. The final prediction model improves prediction accuracy. Experiments show that the random forest algorithm with improved feature selection is optimized, which is compared to other regression algorithms in terms of demand prediction accuracy and fitness. The method is effective.