EVALUATING THE PREDICTIVE POWER OF AN ENSEMBLE MODEL FOR ECONOMIC SUCCESS OF INDIAN MOVIES

Abstract

The Indian motion picture industry has experienced phenomenal growth during the last few decades and plays an important role in emerging economy of India. This paper integrates three analytical models in order to address the intriguing problem of revenue prediction of movies in Indian film industry. The paper attempts to investigate the determinants leading to the success of indigenous movies in Indian context. Ensemble model has been constructed by integrating the three analytical models (Neural Network, Classification and Regression Tree and Robust Regression) using linear optimization approach. Further, a four-way comparative analysis of these three models along with Ensemble model has been carried out. The predictive power of the models has been evaluated using four performance metrics namely root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and large prediction error (LPE). Analyzing novel and original data of 120 Indian movies released during the period August’06-October’15, this paper inspects the nitty-gritties of Indian film industry and seeks to explain the nuances. The study revealed that factors like hype generated on web by a movie, screens on which the movie is released, rating garnered by movie and its genre are the most influential variables in deciding the box-office performance of a movie. Further we observed, that the neural network model closely competes with ensemble model in terms of predictive accuracy. The ensemble model considerably reduces the predictive errors and yields better results on two of the performance metrics.