Classification problems in machine learning

Abstract

In this report, we will introduce and study some different classifiers of scikit-learn which is machine learning in Python. First, we apply 3 different classifications (SGD classifier, random forest classifier and k-nearest neighbors classifier) on a model of identifying a digit from 70,000 handwritten images and compare their advantages and disadvantages. Then we make the combination of the other 3 classifications(logistic regression classifier, decision tree classifier and k-nearest neighbors classifier) as a new ensemble classifier by the majority voting principle. We evaluate and tune the new ensemble classifier through the decision area. And the regularization parameter C of logistic regression classifier and the depth of decision tree are tuned by grid search.