How might we better understand, detect and alert on compromised or suspicious accounts based upon unexpected actions or transactions within environments and applications?
Solution -
We have considered two datasets for implementation (one provided as part of hackathon, one custom) and performed various EDA methodologies to get insights of the fraudulent transactions. As a part of model implementation, we tried One class SVM, IsolationForest, LocalOutlierFactor and LSTM and after seeing the accuracy and performance selected final model. Thus, also created a prototype to implement the Anomaly detection process for related data and get the results as a prediction.