Supervised Learning - Classification Week 7 Challenge

1. Pros and Cons of SVM (Support Vector Machine)

✅ Advantages of SVM:

1. Effective in High-Dimensional Spaces: Works well when the number of features is large.
2. Robust to Overfitting: Especially useful in small datasets where the number of samples is less than the number of features.
3. Works with Non-Linear Data: By using kernel tricks (e.g., RBF, polynomial), SVM can efficiently classify non-linearly separable data.
4. Memory Efficient: Uses a subset of training points (support vectors) rather than the entire dataset.
5. Handles Noisy Data Well: The margin maximization principle ensures a good generalization.

❌ Disadvantages of SVM:

1. Computationally Expensive for Large Datasets: Training can be slow when the dataset is large.
2. Hard to Tune Kernel Parameters: Choosing the correct kernel and parameters requires experimentation.
3. Not Optimal for Noisy Data: Sensitive to outliers since it tries to maximize margin with strict boundaries.
4. Difficult to Interpret: Unlike decision trees, SVMs do not provide direct feature importance.

---

2. Applying SVM to the "Surface Defects in Stainless Steel Plates" Dataset

Now, I'll proceed with:

• Preprocessing the dataset (handling missing values, encoding categorical variables if any).
• Splitting the dataset into training and testing sets.
• Training an SVM model on the dataset.
• Evaluating model performance using accuracy, precision, recall, and F1-score.​

SVM Classification Results for Surface Defects in Stainless Steel Plates:

• Accuracy: 78.4%
• Classification Report:

precision recall f1-score support

0 0.79 0.90 0.84 246

1 0.78 0.58 0.66 143

accuracy 0.78 389

macro avg 0.78 0.74 0.75 389

weighted avg 0.78 0.78 0.78 389

Findings:

• Class 0 (Non-Faulty) has better recall (90%), meaning the model effectively identifies non-faulty cases.
• Class 1 (Faulty) has lower recall (58%), suggesting some misclassification of faults.
• Overall performance is decent (78% accuracy), but improving recall for faulty cases is necessary.

Next Steps for Improvement:

• Hyperparameter tuning (adjusting C and gamma values).
• Trying different kernels (Polynomial, Sigmoid) to compare performance.
• Balancing the dataset if there's an imbalance in class distribution.