Projects‎ > ‎

Digit Recognition

You built a SVM classifier to detect the hand-written digit '9' in Question 3 of Homework 3. Now build a multi-class classifier to be able to detect the digits 0-9. You should compare the performance of KNN, decision trees and SVM (and any other classifiers that you would like to experiment with).

1) Data Extraction:
1.1) Convert the test samples under the Data page into vector tuples (in either python or R)

2) Feature selection and model building:
2.1) Build a supervised learning classifier to detect a digit. You could either train a binary classifier for each digit or train a multi-class classifier
2.2) You should eliminate useless features from the training set.
2.3) Repeat for different classifiers.

3) Classifier evaluation:
3.1) In each case how well does your classifier perform out-of-sample? Provide a 10 x 10 truth table.
3.2) Compare the performance of each classifier.