Red Wine Quality Prediction

Using the UCI red wine quality dataset (1,599 samples, 11 physicochemical features + quality), this notebook reframes sensory quality as a three-class problem — bad (scores 3–4), average (5–6), and good (7–8) — then benchmarks four classifiers after careful cleanup and rebalancing.

Every number below comes from the committed notebook outputs in Red Wine Quality Classification 93 Accuracy.ipynb: Random Forest wins at 92.74% accuracy on a held-out test set of 496 samples.

Raw quality scores are skewed: 3→10, 4→53, 5→681, 6→638, 7→199, 8→18. After IQR outlier removal the frame is 974 × 12. Quality is then binned into bad / average / good, which left a severe imbalance (average 826, good 116, bad 32). SMOTE rebalanced to 826 per class (2,478 rows total) before an 80/20 train–test split (1,982 / 496).

Predictors are the standard UCI physicochemical set: fixed acidity, volatile acidity, citric acid, residual sugar, chlorides, free / total sulfur dioxide, density, pH, sulphates, and alcohol. The original quality column is dropped once new_quality is encoded.

92.74%

85.89%

82.06%

64.52%

Random Forest led with 0.9274 accuracy and balanced precision/recall across classes (e.g. class 1 recall 0.99, class 2 recall 0.96). Confusion matrices and classification reports are saved in the notebook cells alongside the accuracy prints — evidence for the “93%” title on the file.