DBias: Predicting attribute effectiveness using biased databases
Ashwini Dalvi, Irfan N. A. Siddavatam, Abhishek Patel, Aditya Panchal, Faruk Kazi, Sunil G. Bhirud · 2021 International Conference on Smart Generation Computing, Communication and Networking (SMART GENCON) · 2021
With the fast-moving world, the prominence of Artificial Intelligence (AI) algorithms and methodologies has also reached new heights. Modern algorithms are being implemented in each field, including cyber threat Intelligence (CTI). However, multiple frameworks are present in this space, each using its own AIs for more efficient detection. Understanding these algorithms is the key to realise how these algorithms compute each factor, and many Explainable AI (XAI) implementations have been moderately successful in doing so. However, on most occasions, these XAI implementations do not agree with each other when decoding a black-box model. The primary idea put forward in this paper is to determine if the present XAI interpretations of complex AI models have attributes that unfairly contribute more than they should. The paper adopts a novel approach of using biased databases to train multiple models using the same methodology and then simulates them for each scenario. For the current implementation, research was carried out using a popular CTI database.