Supervised Learning - Classification Week 8 Challenge

1. Applying K-Nearest Neighbors (KNN) to the "Surface Defects in Stainless Steel Plates" Dataset

I will now:

• Preprocess the dataset (scaling features, handling missing values).
• Apply KNN classifier with an optimal K value.
• Evaluate performance using accuracy, precision, recall, and F1-score.
• Compare results with SVM to identify differences.

2. Pros and Cons of KNN

✅ Advantages of KNN:

1. Simple and Easy to Implement: No assumptions about data distribution.
2. No Training Phase: KNN is a lazy learner, meaning no model is built before making predictions.
3. Non-Parametric: Works well for non-linear decision boundaries.
4. Works for Multi-Class Classification: Can classify multiple categories easily.

❌ Disadvantages of KNN:

1. Computationally Expensive: Requires storing and comparing all training samples.
2. Sensitive to Noisy Data and Outliers: The performance can degrade if irrelevant features are included.
3. Curse of Dimensionality: High-dimensional data can lead to poor accuracy.
4. Needs Optimal K Selection: Choosing the wrong K value can significantly affect accuracy.

KNN Classification Results for Surface Defects in Stainless Steel Plates:

• Accuracy: 77.89%
• Classification Report:

precision recall f1-score support

0 0.79 0.88 0.83 246

1 0.75 0.60 0.67 143

accuracy 0.78 389

macro avg 0.77 0.74 0.75 389

weighted avg 0.78 0.78 0.77 389

Comparison of KNN vs SVM

Model	Accuracy	Recall (Class 1)	Precision (Class 1)	F1-Score (Class 1)
SVM	78.4%	58%	78%	66%
KNN	77.89%	60%	75%	67%

Key Differences:

• KNN slightly underperforms SVM in accuracy (77.89% vs. 78.4%).
• KNN has a slightly better recall for classifying faulty cases (60%) compared to SVM (58%).
• SVM has a higher precision (78%) than KNN (75%), meaning it misclassifies fewer non-faulty cases.

---

Key Takeaways:

• SVM is more reliable for precise classifications.
• KNN is slightly better for detecting more faulty cases (higher recall for Class 1).
• KNN is computationally expensive for large datasets, whereas SVM handles large data better with kernel tricks.
• KNN is sensitive to the choice of K, so tuning the K value might improve performance.