Bandits using Fuzzy to fill Knapsacks: Smallholder Farmers Credit Scoring

Abstract

Consumer lending is an exercise where risk scoring takes the form of a typical decision making problem. For smallholder farmers, the credit scoring becomes specifically more challenging with the data gaps and outliers in data. Added to that, the process must be as cost-effective as possible while providing as accurate results as possible. This paper uses data obtained from smallholder farmers in Kakamega County in Kenya to set up an experiment of credit scoring as a Bandit with Knapsack problem with Fuzzy Unordered Rule Induction Algorithm (FURIA) being used as the exploit-explore algorithm and Fuzzy Analytical Hierarchical Process (FAHP) used to determine the ranking and consistency of the FURIA rules. The experiment returns a consistency ratio of 0.000529 which is significantly less than the 0.10 threshold. In this regard, the paper proposes the use of FURIA to reduce the regret in Bandit with Knapsack (BwK) as a technique for smallholder credit scoring.