Towards Robust Machine Learning Models for Data Scarcity