Problem Statement:

Smartphone apps are changing our lives and are gaining more and more attention from people everyday. Based on data from statista.com, over 6000 new smartphone apps are produced everyday. However, most of them “died off” before they even get a chance to show up in our smartphone app search engine (e.g., Google Play store). So, how could a cellphone app survive such brutal competetion? Or what qualities must it possess to eventually be installed by customers? In this capstone project, real-world datasets collected from Google Play Store have been analyzed to provide insights to these questions. Different machine learning models have also been developed in attempt to predict the app ratings. The analysis outcome from this project may benefit software/app designers/engineers to adapt the app design in an effort to increase the likelihood of receiving a positive review.

Approach:

Data Wrangling, EDA, Multi-Regression Analysis, Sentiment Polarity and Subjectivity Analysis, Bootstrap Hypothesis Testing, Feature Engineering, Hypertuning Parameters.

Using Machine Learning to Predict App Ratings on GooglePlay Store.

Conclusions:

1. Randomforest classifier has the best prediction for App ratings, with >73% accuracy
2. Number of review, installs and the app size are the critical design features for a higher rating on GooglePlay Store
3. Number of review, installs and the app size are all positively correlated with ratings
4. See more recommendations at the following links.

Code:

https://nbviewer.jupyter.org/github/huanxia1/Predict-App-ratings-on-GooglePlay-Store/blob/master/Predicting%20App%20rating%20on%20GooglePlay%20Store.ipynb#DataWrangling

Report:

https://docs.google.com/document/d/1-0fJv4vlm1-vqEiX8ZaPeoRboqicvYs3klE_TUI0t_E/edit?usp=sharing

Slides:

https://docs.google.com/presentation/d/1GibgdH21hMk9iM_M8ORQQa7kl8Lr0juvYPtEASt4WnU/edit?usp=sharing