Human Activity Recognition – SVM & PCA (Multiclass Classification)

This project applies Support Vector Machines (SVM) and Principal Component Analysis (PCA) to the Human Activity Recognition (HAR) dataset.
The goal is to classify 6 physical activities using smartphone sensor data.

Part 1 – SVM (Without PCA)

• Data preprocessing
• Z-score and Min-Max normalization
• Multiclass SVM (OvO)
• Linear, RBF, Polynomial kernels
• Hyperparameter tuning (C, gamma)
• Accuracy comparison
• Training time measurement

Part 2 – PCA + SVM

• PCA applied to 95% explained variance
• Dimensionality reduction from 561 → PCA components
• SVM retrained on reduced data
• Accuracy comparison (before vs after PCA)
• Training time comparison (before vs after PCA)

Results

• RBF kernel achieved the best accuracy
• PCA significantly reduced training time
• Accuracy remained almost unchanged

Technologies

• Python
• Pandas
• Scikit-learn
• Matplotlib
• Jupyter Notebook