Data-led research projects

In 2019, I presented my paper "Heart Transplant Outcomes Prediction" at Women in Machine Learning (WIML) at Neural Information Processing Systems (NeurIPS) 2019, Vancouver, Canada. First, we used ADASYN to generate synthetic data points that resembled real-life data points to correct the data imbalance in the deaths and survivors (Adaptive Synthetic Re-sampling technique). The Fast Correlation Based Filter method was used to cut the feature set by 80%. We also talked about how urine profiling (a low-cost, non-invasive test) can help predict graft failure early on. My most important takeaway was to ensure that our models and variables were understandable to the healthcare experts.

Previously in 2018 , I interned at Indian Institute of Management, Bangalore, on developing NLP models for root cause analysis of farmer suicides in India. I scraped and analysed the text of 15,000+ suicide articles dating back five years using an automated Selenium-based crawler and word2vec embeddings. The NLTK and GENSIM Python packages were used for unsupervised topic modelling of suicide cases.  It was interesting to see the correlation between topics such as water shortage and nearby cricket fields (60,000L+ of usage daily) However, it was unsurprising to discover a strong correlation between suicides and a drop in crops and government subsidies.