Microfinance Services: Using Machine Learning to Predict Feasibility in Kenya - Code and Paper
During my time at the Chinese University of Hong Kong, I had the opportunity to enhance my data science skills through various courses. This included an advanced economics model where I utilized machine learning techniques to predict the performance of microfinance services in Kenya. My paper investigates usage of informal Kenyan microfinance investment groups, known as chamas, and analyses the feasibility of implementing a web application to assist in chama organisation and management.
For my final paper, I utilized the Kaggle dataset "Islamic Microfinance Services Feasibility Study" by Kinuthia, R. (2018). The data from this dataset, which pertains to microfinance services in Kenya, was used to train predictive models. Retrieved from https://www.kaggle.com/datasets/rkinuthia/islamic-microfinance-services-feasibility-study.
I trained two logistic regression models and a neural network on the survey results to predict chama (informal microfinance investment groups) usage. The first logistic regression model utilized hyperparameter tuning for variable selection, while the second model used p-value significance. The neural network also utilized variables with p-value significance.
The performance of the models was evaluated using confusion matrices and accuracy scores. The first logistic regression model demonstrated high accuracy, suggesting its effectiveness in predicting chama usage. However, the second logistic regression model performed poorly. A neural network analysis was conducted, showing significant improvement in accuracy. The paper demonstrates the potential of using advanced analytics to improve the understanding and use of chama microfinance services in Kenya.
The paper focuses on an unrepresented and vastly unbanked part of potential microfinance clients, the Muslim population. The study shows that belonging to the Christian religion may influence the decision of joining an informal microfinance chama group, and Muslim religion appears to be less significant with a p-value of 0.067. Understanding these factors could help to optimise the financial performance of microfinance initiatives in Kenya and similar contexts. Digitalizing the processes of microfinance in Kenya through a digital platform for managing savings groups is worth exploring further, as it has the potential to make microfinance services more accessible and to create revenue. Making such an app for a sharia compliant population is also worth exploring, as the p-value for the Muslim religion variable is barely not significant.