App Success Predictor

Abstract

A machinery for downloading and extracting features about applications from the Google Play Store was developed and deployed, and the resulting data set was used to train three different models to predict the success of a mobile application; a na ve bayes based text classi er for the de-scription of the application, a generalized linear model which categorizes applications as successful or not, and a linear regression which predicts the average rating of the application. The performance of the models is not su cient to justify their use in driving investments in new applications, however interesting observations about the ecosystem, such as the current trend in photo sharing applications, are elucidated.