Cognitive Analytics and Reinforcement Learning

3Deep Recommender System for Optimizing Debt Collection Using Reinforcement Learning

Keerthana S.¹, Elakkiya R.²* and Santhi B.¹

¹School of Computing, SASTRA Deemed University, Thanjavur, India

²Department of Computer Science, BITS Pilani, Dubai Campus, Dubai, United Arab Emirates

Abstract

Unpaid debt strains the Field Collection Agent (FCA) and causes tremendous financial loss. Most research focused on analyzing the customer credit risk before loan approval to reduce the loss. Few studies focused on streamlining the debt collection process by minimizing the number of collection phone calls made to consumers and predicting the defaulters. None of the articles is focused on field collection agents’ performance and priority allocation. We have framed the objectives such as analyzing FCA’s performance, evaluating the customer’s credit risk, and performing priority allocation by providing personalized recommendations. The FCA’s performance is analyzed concerning different parameters such as collection percentage, number of visits, date and time, and visit intensity. Customer credit risk is analyzed with parameters such as CIBIL score, income, credit history, etc. FCA performance and customer risk categories are frequently changing. Our proposed Reinforcement Learning (RL) model adapts to these dynamic changes and prepares the allocation list, which maps the customer risk category to the corresponding FCA category. In addition, it gives personalized recommendations such as the ...