Final Report for Data Mining class.
Also, participated in corresponding Kaggle competition using this project. Our final and best model PCA + ensemble of neural nets ranked 111 (out of 2696 participating teams and 26000+ entries). Unfortunately, cheating was easy for this challenge, many top teams had perfect score (which means essentially zero misclassification errors, despite having to submit ~34000 individual predictions of integers lying approximately in the interval [100, 700]), which is not very believable, so it is hard to gauge how well we really did.