Using Resampling Techniques with Heterogeneous Stacking Ensemble for Mobile App Stores Reviews Analytics

摘要

Over the past few years, a boom in the popularity of mobile devices and mobile apps has appeared. More than 205 billion apps were downloaded in 2018. Developers directly distribute mobile apps to end users via a centralized platform like the "App Store" for iOS or the "Play Store" for Android. The Mobile app developers get continuous feedback from users' reviews added to these stores. Tools like CLAP or AR-MINER were used to crawl reviews from the stores and try to classify them into lots of classifications like (Bug, required feature, usability issue, performance issue,) to facilitate the categorization of issues or features addition to the developer. Some machine learning techniques is used to get the most accurate data classification to help the developer to classify the reported reviews on the stores. This paper presents a machine learning model that uses the Resampling techniques for handling imbalanced classes in addition to ensemble learning and stacking. The model outperforms those tools and enhances the results applied on Mobile App Stores Reviews Analytics. In addition to that the paper provides experiments applied on different kinds of datasets and showed improvements in accuracy from 85% (previous model) to 90% (our model) and ROC from 96% to 98% especially on the Reviews dataset.