Predicting movie success based on pre-released features

Movie making is a billion-dollar industry. Every month hundreds of movies get released and earn millions of dollars in revenue. However, majority of the movies fail to create an impact on the Box-Office and flop. This not only put a bad impression on the entire cast and crew but also creates a huge setback in financial terms. As a producer or investor, it is crucial for them to have some certainty that the money they are investing in will give a good return otherwise they'll lose all their capital eventually. The idea of this research is to predict based on certain pre-released variables of the movie, whether an upcoming movie is going to succeed or fail in monetary terms. Many researches have already been doing that in this domain based on different techniques and around different datasets. The novelty of this research is that the proposed approach is not only based on classical movie features, but incorporates all other dependencies as well such as star power, popularity of the cast, track record of director, and actors, to predict whether movie will succeed or fail and whether an investor should invest in the movie proposal or not. This article uses multiple machine learning algorithms and tested them over various evaluation metrics. Among them, CatBoostRegression and Stacking Regression outperformed the remaining by giving the maximum model accuracy of 83.84% and 83.5% respectively. The article have used IMDB Movies Extensive Dataset. This dataset contains information of movies from 1894 to 2020 and has at least 100 votes.