Can the attributes of an app predict its rating?

Abstract

When Google Play Store developers first publish their apps, it may seem that their success comes from the roll of a dice. Ways to improve the rating of their app may seem impossible to glean, with many relying on blind guessing to determine ways forward. In this paper, we hypothesize that the attributes of a Google Play Store apps, including maturity level, install count, and price, can estimate its rating, or stars out of five. We believed that comparing three models trained on these attributes, each with unique architectures, would lead to a higher final accuracy than one. By using Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) classifiers, we found that the attributes of an app can predict its rating, with review count, date of last update, and storage size being the most influential attributes. During modeling, we found that the RF classifier most successfully predicted the rating of an app, getting 79.3% of predictions correct. These results suggest a connection between the rating of an app and its attributes. The results from this paper can inform app developers and investors about improvement paths to increase the rating of their app, thereby increasing its success.