– Built binary image classifier with SVMs using convex optimization solver CVXOPT Exploring Gaussian kernels

– Extended SVMs to multi-class classification using one-vs-one classifier and Visualized frequent miss classifications from confusion matrix, Performed Hyper parameter tuning with 5-fold cross-validation to find optimal C parameter

– Achieved validation set accuracy of 66% in dataset containing 2380 training images of size 1501503 with 6 classes