In this thesis, we look at the extent to which the rating of an application can be predicted using only its publication strategy. The goal is to determine the importance of this latter in the way a customer perceives an application. The following classification algorithms are used to predict the rating: decision tree, random forest, KNN, SVM and MLP. We use metadata about 96,178 Android apps and 81,000 IOS apps. The performances of the algorithms are compared and discussed. We conclude that, for both stores, 50% of the ratings can be predicted using only variables controlled by the companies before the commercialization and displayed on the store. Moreover, the most important variables for the predictions are the length of the description, the size, the length of the title and the number of screenshots. Some recommendations are made to apps designers and future works are discussed.
|Date of Award||17 Jun 2020|
|Supervisor||Stephane Faulkner (Supervisor)|