Selected work that I'm proud of.

ML pipeline scoring 7,043 telecom customers by churn risk across four models. XGBoost wins at 98.1% recall, saving €5,620 per cycle by treating a missed churner as 20x costlier than a false alarm.

Replaces manual Excel forecasting (30% systematic under-forecast) with a per-segment ML pipeline. Best model selected per segment via 5-fold CV, delivering 3-month forward forecasts with 80% and 95% confidence intervals.

Segments 1,607 active customers into four groups (Strategic, Anchor, Rising, Base) using K-Means, validated against RFM tiers. Identified 311 high-potential accounts never yet called on, representing $851K in untapped revenue.

Four-page Power BI dashboard covering $24.9M revenue, 25,200+ orders, and 17,400 customers across 10 territories. Includes 23 DAX measures and a real-time price adjustment tool for modelling pricing scenarios.

Analysis of the WHR 2018 dataset (131 countries, 6 indicators) using PCA, K-means clustering, and OLS regression. PC1 alone explains 50.8% of happiness variance. Completed at Aalto University.