multi-target_pred

Utilizing a dataset of 10,000 mobile apps hosted in the GooglePlayStore, plot models identify variables determinant of application success and evaluate the relative feature importance in multiple-target regression models to optimize their prediction accuracy of application rating on a decimalized 1-5 scale.

main analysis, model prediction

• Scatterplot Distributions of Rating v Reviews, Size, Installs by Type
• Prediction Analysis of App Rating using Random Forest Regressor (RFR), Gradient Boost Regressor (GBR), Decision Tree Regressor (DTR), AdaBoost Regressor (ABR) base models.
  • Evaluating Average Error, Accuracy, Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) of models
  • Plotting Model Predictions of Rating vs Actual Rating based on Reviews, Size, Installs
  • Plotting Prediction Accuracy of RFR, GBR base models and DTR, ABR base models to Test Set
• Hyperparameter Tuning of RFR, GBR models using GridSearchCV, Feature Importance evaluation of best grid using SHAP Tree Explainer
• Stacked Generalization ensemble (RFR, GBR, XGBR, RidgeCV estimation) comparison to RFR, GBR base model accuracy

feature correlation

• Heatmap Correlation of one-hot encoded Categorical Features
• K-means Cluster Evaluation for binning Reviews, Size
• Standardized Scaling v Normalized Scaling for Reviews, Size, Installs

base linear models

• Linear Regression Analysis of Reviews, Size, Installs on Rating for each App Category

tier segmented linear models

• Linear Regression Analysis of Tier Segmented Reviews, Size, Installs, Type, Content Rating on Rating

weight encoded categorical feature evaluation, pairwise plots

• Weight of Evidence (WoE) encoding, Information Value (IV) of Categorical variables (Categories, Type, Content Rating)
• RFR Model Evaluation of Rating Prediction based on Reviews, Size, Installs, WoE Categorical variables
  • Evaluating Feature, Permutation Importance determined by RFR Model using SHAP TreeExplainer
• Pairwise Plots for Categorical Feature Evaluation using WoE-encoded Categorical variables
• Target Enclosed Feature Modeling based on Example Rating

jupyter notebooks running python, pandas, numpy, matplotlib, sk.learn