Classification experiments (KNN, SVM, decision trees, and a hyperparameter-tuned random forest) on the combined UCI Wine Quality dataset (red + white), predicting wine quality (multiclass, 3–9) from 11 physicochemical attributes.
winequality-red.csv and winequality-white.csv — Vinho Verde wine samples from Cortez et al. (2009). The two are concatenated with a binary type indicator column and modeled jointly as a single multiclass quality-prediction problem (stratified 80/20 train/test split, seed=42).
📖 Full results table (incl. macro-F1) and evaluation methodology → project Wiki
flowchart TD
Red["winequality-red.csv"] --> Combine["main.py — concat + type flag"]
White["winequality-white.csv"] --> Combine
Combine --> Split["stratified train/test split (80/20)<br/>+ StandardScaler"]
Split --> Baseline["Baseline: KNN / SVM / Decision Tree"]
Split --> Tuned["GridSearchCV-tuned:<br/>Random Forest / KNN / SVM / Decision Tree"]
Baseline --> Tables["Tables/ — classification reports"]
Tuned --> Tables
Baseline --> Plots["Plots/ — confusion matrices<br/>+ feature importance"]
Tuned --> Plots
python main.pyRequires numpy, pandas, matplotlib, seaborn, scikit-learn. Writes classification reports to Tables/ and confusion-matrix / feature-importance plots to Plots/. Paths are resolved relative to the script location, not hardcoded.
| Model | Accuracy |
|---|---|
| KNN (baseline) | ~0.56 |
| SVM (baseline) | ~0.53 |
| Decision Tree (baseline) | ~0.60 |
| Tuned Random Forest | ~0.69 |
| Tuned KNN | ~0.67 |
| Tuned SVM | ~0.59 |
| Tuned Decision Tree | ~0.60 |
Exact figures vary slightly run to run depending on GridSearchCV grid choices; see Tables/classification_report_*.txt for the current run's numbers.
Classification accuracy alone is reported here; per repo conventions, any follow-on analysis reporting model comparisons should include effect sizes/confidence intervals rather than point accuracy alone. Note the class imbalance (qualities 3 and 9 have single-digit support) means macro-averaged metrics are far more informative than raw accuracy for the tails of the distribution.