Density Estimators for Positive-Unlabeled Learning

IE 506: Machine Learning: Principles and Techniques (Course Project)

Under the Guidance of Prof. P. Balamurugan 

• Implemented Generative PU Learning with Bayesian Network as the generative model on categorical data with 1-hot encoding.

• Derived 3 experimental settings for 7 datasets, conducted 10-fold CV, and used the K2 scoring function to avoid overfitting.

• Applied SVM with RBF kernel and Random Forest for classification, achieving average accuracies of 75%, 78% respectively.