preview

The Importance Of Classifiers

Decent Essays

After removing the useless attributes, we are left with 640 attributes. When we retrain the Naive Bayes and J48 classifiers using this reduced dataset, we got the same accuracy as we got in A(d) that is before reducing the attributes. The accuracy of Classifiers on the reduced dataset: ⦁ Naive Bayes: Accuracy = 70.85% ⦁ J48: Accuracy = 72.95% As we can see the accuracy of both the classifiers is intact even after removing many attributes. Here we can observe that performance of both the classifiers remains same, it neither improves nor degrade, this proves that the removed attribute didn't contribute to the performance of both the classifiers and didn't help classifiers in classifying the instances. Thus, such dimensions should be …show more content…

⦁ As we can see the difference in plots of the attributes against the class variable for the highest ranked attributes(for example, pixel_14_15) and the lowest ranked ones(pixel_2_11), in the former plot we can see that if we select this attribute and split it using some value range then we will get more meaningful sub-datasets and we can use those sub-datasets to classify the instances however in the second case if select that attribute and try to split it, we will get impure data and we won't be able to classify data as almost all instances have similar value for pixel_2_11 attribute. Here, the first attribute has high information gain however the second attribute has 0 information gain. ⦁ It's not safe to remove all the attributes with Information gain as 0. We should remove the attributes with IG 0 and run the classifier to make sure that performance has not degraded by removing those attributes. For example, In classifier's result we can see the 'Incorrectly Classified Instances', we started removing the attributes with IG 0 and retrained the classifier with new dataset and checked the value of incorrectly classified instances to make sure that that value does not increase and if it does and even after removing more attributes it does not come back to global low value then we should retrieve those attributes back and remove only those attributes when we had lowest Incorrectly Classified

Get Access