Kaan Kartal Kuyucu
I'm drawn to the strategic side of business — where to allocate resources, which risks are worth taking, which customers actually matter. Data and modeling are tools I use to get sharper answers to those questions, not the headline.
Projects
Satellite-Based Land Use Compliance Screening
CNN-based land cover classification on Sentinel-2 imagery, simulating an EU-style agricultural subsidy compliance tool.
Python, TensorFlow/Keras, EuroSAT (Sentinel-2)
Telecom Churn: From Prediction to Retention ROI
Improves on a public Kaggle churn baseline (fixes data leakage and multicollinearity, beats its 77% accuracy on every metric), then adds an expected-value retention-targeting layer most churn projects skip — backed by a live, interactive dashboard.
Python, scikit-learn, XGBoost, Pandas, React/Recharts (live dashboard)
Credit Card Fraud Detection
An XGBoost fraud classifier on 1.85M simulated credit card transactions — precision/recall-focused evaluation on a dataset where only 0.58% of transactions are fraudulent.
Python, pandas, XGBoost, scikit-learn, Jupyter
Favorita Store Sales: 15-Day Demand Forecasting
A LightGBM demand forecaster for Corporación Favorita's 54 stores × 33 product families, beating a naive "last week's value" baseline by 31.9% (RMSLE 0.3878) — validated the way the competition actually scores, then extended into a real Kaggle test-set submission.
Python, LightGBM, pandas, scikit-learn, Plotly
Audi A1 Price Prediction
Linear, Ridge, and Lasso regression on 471 real UK Audi A1 listings, with SHAP and confidence-interval explainability, deployed as a live price calculator backed by an Azure Function API.
Python, scikit-learn, statsmodels, SHAP, Azure Functions