Predict App Success on the Google Play Store using Machine Learning - Decision Trees

UCSD-Edx Python for Data Science MicroMasters Final Project

Dataset: Kaggle Google Play Store Apps https://www.kaggle.com/lava18/google-play-store-apps

Motivation:

• Gain edge over the industry competition for app success.
• Provide insight for advertisement companies on which apps would generate the most revenue if ads were added.
• Assist Android developers to develop state-of-the-art apps that the public deserve.

Research Questions:

1. Can you predict an app's popularity on the Google Play Store using a Decision Tree?
2. If a developer were to create a new app, what qualities should this app have in order to generate the most ad revenue?

Results

• ~95% Accuracy if using Reviews and Ratings columns.
• ~72% Accuracy if not using above columns. (More realistic as reviews/ratings will be null if app is not on the Play store.)

UPDATE - Fix These:

Issues found with the model/process used in this Jupyter Notebook:

• When preprocessing the model, OneHotEncoding should have been used rather than the LabelEncoder. This is because a bias may exist between numbers in 0-33 rather than a matrix of 0 and 1 for category, genre, and content rating.
• A K-fold Cross Validation set should have been used to fine-tune the model before using the test set.
• Other models could have been used during the cross validation process to determine the best binary classifier model to use.
• ROC AUC or precision/recall curves should have been created to determine the best decision threshold for the model.

Future Work

• Models which will most likely have better performance are K Nearest-Neighbors, Logistic Regression, and Random Forests. Once implementation of these models are learned, I will commit a change to this repo with the new results from all listed ML models.

Requirements

Jupyter Notebook may be used with any OS. Ensure you have pip installed the latest versions of the following:

• Pandas
• Numpy
• Seaborn
• Sklearn
• Plotly
• Matplotlib