Approximately 8.9 million people, or 17% of Kenya’s population, live below the poverty line of 1.9 USD on a daily basis, majority of them in the rural areas. This research aimed to model the impact of self-help groups on livelihoods in rural areas of Kenya using machine learning techniques. To achieve this, the study employed Logistic regression, Naïve bayes algorithm and Support vector machine and compared their performance using various metrics to find out whether or not one of the three models outperformed the other two models. The study used principal component analysis to evaluate the most significant factors that affected group performance with an aim to ensure that the self-help groups achieved their original set objective. The study used primary data obtained through the issuance of structured questionnaires to self-help group members, on their wealth status since joining the self-help groups on areas such as ability to save, access to credit services and acquiring assets, both income generating and household assets. The target population was 2250 and the sample size was obtained using the Yamane's formula. The study’s findings helped identify the key predictors of members’ livelihoods and provided insights into how self-help groups influence them. The research also explored the factors that contributed to the success or failure of self-help groups. The results of the principal component analysis indicated that the frequency of the meetings and the quality of discussions held was the factor that contributed the most to group performance, as well as access to credit facilities. The results of logistic regression indicated that 90.9% of the members had seen a significant improvement on their wealth status since joining self-help groups and the significant predictor variables were income generating assets, access to basic commodities and access to loans. The model’s accuracy was 88.04%. The SVM model classified 93.98% of the members as having an improved wealth status since joining SHGs with an accuracy of 84.62%. Finally, the naïve bayes algorithm classified 87% of the members as having a positive improvement on their wealth status with an accuracy of 92.34%. The results of this research are valuable to NGOs and practitioners in determining the effectiveness of self-help groups in promoting sustainable livelihoods and shaping future interventions on SHGs as well as on group members on how best they can make their groups more effective. Furthermore, the study’s results provided more knowledge into the use of Machine learning methods in performing classification tasks and in identifying which method performed best in this specific task. This research contributed to the existing body of knowledge on the impact of self-help groups on rural livelihoods and the use of SHGs as a tool that can be used to reduce or completely eradicate poverty in the lives of people living in rural areas of Kenya. The comparison of the three models to find out which among them performed the best in the classification task was also a significant contribution to the existing body of knowledge. Naïve bayes algorithm outperformed the other two models.