/iris-project

iris project

Primary LanguageJupyter Notebook

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:

Iris 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.