Application of Ensemble Machine Learning and Big data to help Financial Institutions detect payment fraud & prevent crime losses and reduce false positives.

Financial institutions are under tremendous pressure due to rapid changes in fraud landscape. FIs are majorly facing dual challenges: financial damage and loss of customers trust & shareholders value. The rapid advancement in technologies has enabled fraudster with easy access to information which is being leveraged for committing crime. In the recent past, data scientists have been using several predictive models and machine learning algorithms like Logistic Regression, decision tree, Neural Network, Support Vector machines, etc. to predict fraud. However advance feature engineering has not been emphasized enough to build a predictive algorithms. Too many features can create complexities while predicting fraud, create over fitting problem and also hamper model accuracy. We would be applying a novel approach for feature engineering to select right set of optimized features and build robust & consistent algorithms. To prevent fraud & crime losses and improve customer experience, we have proposed an ensemble approach of Generalized Linear Model and Gradient Boosting Models. Gradient Boosting Model has been built for predicting errors produced by GLM and then later a standalone GBM model has been built. Error predictive score was ensemble with Stochastic GBM to avoid overfitting, reduce variance and improve accuracy.