Applied Machine Learning
Four problem types, one discipline: baseline first, compare before choosing, then explain the choice.
- Context
- Machine Learning coursework, individual
- My part
- Everything: four notebooks and their write-ups
- Stack
- Python, scikit-learn, statsmodels, pandas
- Year
- 2025
0.875
F1 on faulty machines, a class that is 3.39% of the data
- What it is
- Factory fault classification, house-price regression, customer segmentation and monthly gas, electricity and water forecasting.
- What I did
- I scored the imbalanced fault model on the minority class, not accuracy; compared KMeans, GMM and Ward clustering with repeated-seed stability checks; and forecast on a strict chronological split, choosing between ETS and tuned SARIMAX on validation error before a 60-month forecast.
- Result
- The forecasting and segmentation notebooks are the strongest; the house-price model is honest about being modest (R² 0.50).
