Churn Prediction for KKBox
Predicting subscriber churn for KKBox’s music service, comparing five classifiers and landing on interpretable logistic regression with model monitoring.
Recent posts
Human Action Recognition in Time Series data using Deep Learning
Classifying six human activities from raw smartphone sensor data with LSTMs and 1D CNNs, and why 1D CNNs win on runtime for production.
Hazard Count Prediction for Liberty Mutual’s Properties
A top-12% Kaggle finish predicting property hazard counts for Liberty Mutual, where ensembling and disciplined cross-validation beat feature engineering.