An Application of Ensemble and Deep Learning Models in Predictive Analytics

Predictive analytics stands for prediction of future demand with the help of historical data. An accurate prediction of future demand is the need for a competitive and challenging market. Thus a robust and accurate model is required to cater to such needs. This paper proposes two novel approaches for taxi demand prediction that is about to come in the next ten-minutes interval. The first model is a three-stage ensemble model for prediction where the first stage consists of data cleaning and exploratory data analysis (EDA) and then Mini-Batch K-Means clustering is applied, followed by time-binning in the second stage and in the final stage the processed data is fed to the Ensemble Model. The second model is a four-stage deep learning model where the first two stages are common as the proposed ensemble model. In the third stage, the Fourier Transform of the time series data is used to extract and add more features to the training set and in the final stage, the processed training set is fed to the deep neural network (DNN) for prediction. The NYC TLC trip dataset is used for the experiments, and the results clearly indicate that the proposed models perform better than the existing models and the deep learning model performs the best among all.