A Comparative Study of Reinforcement Learning Algorithms for Distribution Network Reconfiguration Wi

Credit Defaulters in Banking is a term used for account holders unable to repay financial loans in time. Till recently, credit and loan systems were human dependent and had limitations in identifying defaulters due to weak data monitoring. Today, in the era of Data Science and Machine Learning (ML), data is closely monitored with inferences and timely feedback. However, the data available for credit defaulters is usually heavily imbalanced. The objective of this paper is to use a technique named ‘Reinforcement Learning’ (RL) to mitigate this bias using the concept of reward feedback wherein current decisions influence future decisions. We aim to do a comparative study of the Double Deep Q-Network (DDQN) algorithm under reinforcement learning with the existing supervised learning algorithms and find out if RL is suitable for the task. We find that RL gives a performance at par with supervised learning, in dataset specific cases.