Predictive Analytics using Ensemble Models

摘要

Predictive Analytics means forecasting future demand using historical data. A robust, accurate and fast prediction model is the need of the present constantly changing competitive environment. More specifically, for taxi demand prediction several statistical and regression-based models exist, but still, the accuracy is a challenging issue. This paper proposes a three-tier ensemble model for more accurate and faster taxi demand prediction. In the first stage, data cleaning is done followed by Mini Batch K-Means Clustering in the second stage whereas the Random Forest (a bagging technique) is applied at the last stage for demand prediction. The NYC TLC trip dataset is used for experiments, and the results obtained highlight that the proposed model outperforms the statistical models.