• Algorithm Development: Improving multi-label classification by generating new data for rare labels @TCS Big Data Lab, Rajasthan

Problem: Tail-labels are labels that have very few instances in the training data, making it hard for multi-label classifiers to learn them well.

Solution: A data augmentation technique called MLSMOTE is used to create synthetic instances for the tail-labels and add them to the training data. Then, a deep learning model called LLSF-DL is trained on the augmented data to improve the multi-label classification performance.

• Paper Implementation : Learning Label-Specific Features for Multi-Label Classification (LLSF)

First Python implementation

- LLSF is a novel method for multi-label classification that leverages label-specific features.

- This algorithm can also perform feature selection for multi-label learning by ranking the features according to their relevance to each label.

- We can enhance existing multi-label classification algorithms that use binary classifiers by applying LLSF to each label separately and combining the results.

#Electricity price prediction using Extreme learning Machine, PSO and ARIMA hybrids.

To capture frequent changes occurring in the electricity prices, we introduced hybrids of Extreme learning machine and Particle-Swarm-Optimization to achieve even greater accuracy.

#Session-based Recommendation with Graph Neural Networks.

The project focused on learning from graphs using Graph neural nets – GNN, to capture essential innate features to recommend items during an ongoing session.

#A novel SVM-kNN-PSO Ensemble Method for Intrusion Detection System

A research paper summary produced at behest of the Soft-computing supervisor Dr. NIstha Kesswani.