Video Action Classifier

Binary and multi-class models built to classify video actions for CS 74, Machine Learning. The approach leveraged precomputed features to run time series and SVM models over 20 frames for each video to train the classifiers. A Bayesian hyperparameter optimizer was designed to tune the models.

Achieved the highest test accuracy in the class — 100% for the binary classifier and 96.6% for the 30-class multi-class classifier — outperforming 200 CS students in both competitions and beating the next best scores by 10%.