Design of contextual agricultural credit scoring model (CACSM) for Kisan Credit Card scheme in India: An application of NLP and machine learning

This study introduces a contextual agricultural credit scoring model (CACSM) that incorporates loan repayment intention scores to enhance decision-making in collateral-free, small-ticket agricultural loans. By integrating sentiment analysis using Valence Aware Dictionary and sEntiment Reasoner (VADER) with traditional credit scoring variables, the model aims to improve loan approval outcomes. The development of CACSM involves three key stages: first, determining the optimal alpha parameter for VADER sentiment analysis to assess repayment intentions; second, evaluating the performance of machine learning models using traditional credit scoring factors; and third, analysing model performance when loan repayment intention scores are added. Experimental results show that incorporating repayment intention scores significantly improves predictive accuracy, with both the decision tree classifier and gradient boosting classifier achieving high F1 scores (0.956522) compared to traditional models (0.454545). This approach can enhance credit officers’ judgment in lending decisions for the Kishan Credit Card scheme in India, offering potential scalability with further robustness testing.