SVM

Spam mail detection using SVM

Use SVM to classify emails into spam or non-spam categories and report the classification accuracy for various SVM parameters and kernel functions.

Dataset

• spambase.data :
  • Number of Instances: 4601 (1813 Spam = 39.4%)
  • Number of Attributes: 58 (57 continuous, 1 nominal class label)
  • Class Distribution: Spam 1813 (39.4%) Non-Spam 2788 (60.6%)

More info

Code walkthrough

1. Importing libraries
2. Reading dataset and preprocessing
   • Null checking
   • Class symbol conversion (1,-1)
   • Train-Test split (70:30)
   • Normalize based on mean and variance of train split
3. Model building
   • Class for SVM with functions for training and testing
   • params: kernel function (linear, poly, RBF) , soft margin constant
   • methods:
     • fit(X,y) : Solves dual equation of SVM and stores weights and bias of separator
     • project(X) : To project data points using obtained weights and bias
     • predict(X) : Sign function to specify class label
4. Model training and testing
5. Comparing with sklearn library function
6. Visualizing decision boundaries by performing PCA on data

Steps to run the code

1. Install Jupyter Notebook or use Google Colab.
2. Open the file ML_Assignment_2.ipynb in Jupyter Notebook or Google Colab.
3. Run all the cells.

Libraries used:

• pandas
• numpy
• matplotlib
• sklearn
• seaborn

Results (Accuracies)

1. Our SVM results

Kernel	C1	C2	C3
	1	10	100
linear	0.923968	0.923244	0.919623
poly	0.902969	0.897900	0.898624
rbf	0.837075	0.849385	0.847212

2. Scikit learn results

Kernel	C1	C2	C3
	1	10	100
linear	0.923968	0.923244	0.918899
poly	0.843592	0.924692	0.915279
rbf	0.934830	0.934106	0.920348