Deep Learning-driven Explainable AI using Generative Adversarial Network (GAN)

In today's digital world, human decisions are influenced by machine driven AI systems and it is essential to trust the outcome of such systems to justify the decisions whereas, ethical issues in fairness and interpretability of such systems are some of the major risks associated with machine learning models which are developed and trained with biased datasets along with sensitive attributes like race, caste, geographical location, sex etc. Biasness in AI Systems erode trust between humans & machines that learn together, and such biases may compound over time. There is hardly any AI System and machine learning models are completely unbiased or universally fair. Transparency and fairness in the predictions of such models is of a major concern across various business domains. Sensing a big void due to lack of innovative solution to address such problem, proposed solution evaluated various approaches and finally proposed an adversarial Neural Network based machine learning method which not only evaluate the fairness of the model against sensitive attributes (i.e. age, race, gender, sex and so on) to help mitigate the biases without losing much accuracy but improve transparency through interpretability of the model which clearly explain how a particular model arrive to a certain decision.