Authors - Farai C. Jonha, Arthur Ndlovu, Mainford Mutandavari Abstract - This study presents a systematic review on the use of deep learning and density-based techniques for explainable segmentation of banking customers. We analyze 71 peer-reviewed papers published between 2015 and 2025 to investigate their methodological trends, validation approaches, and the degree of incorporation of interpretability into proposed models. Our findings suggest that autoencoders and variational autoencoders provide better separation of clusters than models using raw data. In terms of clustering methods, density-based clustering algorithms perform better than clustering algorithms based on centroids since banking data exhibit highly skewed and non-Gaussian patterns. We also observe a common deficiency in explainability, with less than 26% of the re-viewed papers considering approaches such as SHAP or LIME. Furthermore, considerations of external validity, operational governance, regulation, and scalability of implementation are rare. We therefore propose an explainable customer segmentation (XCS) framework based on deep representation learning, density-based clustering, post-hoc explainability, and an operationally ready pipeline that is suitable for use in regulated banking environments
Open Zoom