XG Boost-driven feature selection for Paygo loan default optimized using hybrid meta-heuristic algorithms

Customer default is a persistent challenge impacting the loan repayment sector, particularly in the Pay-As-You-Go within Africa's renewable energy sector. In off-grid communities, renewable energy companies offer Solar Home Systems, where payments are made incrementally over time, using mobile money daily. Identifying potential defaulters early is essential for these companies' sustainability and profitability. Therefore, there is a pressing need for advanced prediction techniques to address these challenges. The primary goal of this research was to develop a hybrid meta-heuristic model that offers higher predictive accuracy in forecasting loan defaulters compared to traditional classifiers. The use of the Xgboost algorithm for feature selection where parameters of Xgboost algorithm are tuned to achieve optimal parameter points rather than default parameter points, while meta-heuristic optimization used was random forest optimized using Particle Swarm Optimization Algorithm. This meta-heuristic approach aims to achieve higher predictive accuracy by leveraging the strengths of individual classifiers. The research results show that tuning the Xgboost algorithm parameter points to their optimal points in feature selection achieved a significant improvement in feature selection, with prediction accuracy reaching 81.289% to 93.456%. These findings provide valuable insights into developing accurate and reliable Paygo prediction models. The hybrid model's improved accuracy suggests that it is better equipped to handle the complexities of loan default prediction in off-grid communities. Ultimately, this approach can facilitate the wider adoption of solar companies in Africa by mitigating the financial risks associated with loan defaults.