Thesis Title: Fuzzy logic-based multi-criteria fair aware risk scoring model for small scale farmers

Student’s Name: Benjamin H. Otieno

Supervisors Names:
1. Prof. Franklin Wabwoba
2. Dr. George Musumba

Abstract:

The role financial service providers’ play in enhancing productivity by small-scale farmers cannot be over-emphasized. Limited access to financial services is one of the major challenges for small-scale farmers. Unfortunately, financial institutions tend to shy away from this group of clients. One of the major reasons, is the perceived high-risk profile of small-scale farmers Additionally, it is very difficult and sometimes expensive to perform risk scoring for the small-scale farmers. To ensure the small-scale farmers get access to financial services at affordable rate, the lenders and insurers ought to have confidence that the tools and techniques they are using, do not end up exposing them to large non-performing loans (for lenders) and high loss ratios (for insurers). The conventional risk scoring models used for consumer loans and large corporate firms and mortgage loans have assumptions that may not be applicable to small scale farmers. Most small-scale farmers do not have title deeds, do not keep books of accounts, have limited access to collateral and do not have guaranteed markets. Consequently, the financial service providers may to some extent be justified in classifying them as high risk since the actual risk may be unknown in some cases. Using three different datasets, this research experimented various algorithms used in risk scoring and established that fuzzy logic based multi criteria decision making algorithm, outperformed others in accurate and fair classification of loan seekers’ risk profiles. Using a sample of 49 households and 14 financial institutions drawn from Kakamega County, the research defined 14 data variables that could be used in risk scoring. Eleven of the variables were either not easily available or difficult to verify thus uncommon in financial institutions usage thus termed alternative data. Design Science was used to propose a fuzzy logic based and bandits with knapsack model for fair risk scoring using the alternative data. The model developed was then tested using synthetic data and the outcome of the tests, based on the fuzzy analytical hierarchical process, was deemed to be successful hence the fitness for use of the model. The research theorizes that the enhancement of the accuracy and fairness in risk scoring models can boost the confidence of lenders and insurers in dealing with small scale farmers, resulting in increased financial inclusion. Increased access to financial services, could spur increased food production, resulting in enhanced food security.