Automated Machine Learning For Outlier Detection

Outlier Detection has been a fertile area of Machine Learning research for nearly two decades, leading to an explosive growth in the number of algorithms. Each algorithm is unique in its approach and with such a wide pool of choices, decision making usually becomes hard and cumbersome. Often times it requires Machine Learning expertise to select the right algorithm and tune it as per the dataset. While this seems like a minor inconvenience for researchers, it is a significant challenge for novice users and domain experts like doctors, engineers etc. In this thesis we propose to democratize Automated Machine Learning for Outlier detection making it accessible and usable for both beginners and advanced users. We do this by selecting an existing open-source package and building a robust AutoML module into it, which can handle algorithm selection as well as hyper-pramater tuning. Next we build an Interactive User Interface which acts as a one stop solution for a novice user to perform anomaly detection on custom datasets without significant coding efforts. The interface allows for the user to upload a dataset and in return displays an interactive visual representation of anomalies found within the dataset. Unbeknownst to the user, this interface invokes our autoML module in the backend to detect the anomalies This end-to-end functional flow without any interaction with code elements improves the user experience thereby accommodating a wide range of audienc