Anomaly Detection
Solution Overview
Risk Edge’s Anomaly Detection (AD) Solution enables Internal Audit teams to go beyond their rules based approach and use AI to their advantage. AD allows users and auditors to go through every banking / transaction / financial / network entry, and detects if they deviate from a normal behavorial pattern. With the help of Advanced Deep Learning and Machine Learning Algorithms, AD’s AI Engine is able to detect old and emerging patterns in transactions and flag those that deviate from normal behaviour. The solution can be used to detect anomalies in all kinds of data – from banking transactions (Anti-Money Laundering) to credit cards, accounting, cyber-security and trading back-office transactions.
APIs & Infrastructure
Anomaly Detection Engine is capable of connecting with various source systems apart from reading in data from flat files or in json / xml formats. JEAD can be deployed on any cloud environment – AWS, Azure, GCP. We use high-end computation machines to train and test data which are deployed as a pay-per-use service on the cloud and enable users to run their new transactions in almost real-time and see the results on their dashboard.
Automated Data Analysis
Pre-processing of all your transactions is completely automated. Various in-built rules ensure that only qualified, validated transactions make it to the training / testing stage – giving this solution more reliability than the other BI / Analytics solutions running on top of your data warehouses.
Deep Learning Algorithms
Advanced self-learning AI algorithms are used to train and test the models. What’s more, key parameters are exposed via UI so that the users can tweak the algorithms based on the results. The models for training the data can be triggered right from the system Interface, and once done, are available to test the real-time or latest transactions from various systems. Our models are pre-trained and optimized for performance (using parallel processing) and accuracy, so you can get a head-start in identifying anomalous transactions.
Utilizing Multiple Models
Anomaly Detection doesn’t just rely on Deep Learning models, but a host of other Machine Learning and heuristic models to arrive at the output. A broader, more sophisticated framework sits atop the individual models and decides which results should be picked from which model, depending on the past patterns. This not only allows us to re-use existing corporate knowledge inside the system, but also brings in flexibility to use the strengths of some other models to improve the results by a significant margin.