Analyzed 25,331 skin lesion images (taken from Kaggle) which was the ISIC challenge in 2019 as part of a group course project (STAT 557: Data Mining I) with Aditya and Sudhir in Fall 2022.
- Implemented CNN (in R) with 80% test set accuracy to classify all the 25,331 Skin Lesion Images.
- Implemented Synthetic Minority Over-sampling TEchnique (SMOTE) in R to address the heavy data imbalance observed in the dataset.
- Analyzed the weakness of standard techniques like bagging, boosting, random forests, and SVM over CNN.
- Implemented convolutional autoencoder(in TensorFlow) to perform dimension reduction.