The goal of this project is to get used to the Scikit-learn SVM API. I also add XGBoost algorithm ( booster: gblinear
) for comparison.
Dataset:
This project shows the difference of the decision surfaces for difference kernels.
Comparison of different linear SVM classifiers on a 2D projection of the iris dataset. We only consider the first 2 features of this dataset:
- Sepal length
- Sepal width
The accuracy of the chosen algorithms are as follows:
0.8 # 'SVC with linear kernel',
0.6888888888888889 # 'LinearSVC (linear kernel)',
0.7777777777777778 # 'SVC with RBF kernel',
0.7777777777777778 # 'SVC with polynomial (degree 3) kernel',
0.5111111111111111 # 'xgboost with gblinear'
In this part I added ROC-AUC curve with 6-fold cross valication, and calculated the AUC (area under the curve) of iris dataset. The figure roughly shows how the classifier output is affected by changes in the training data.