Development of Hybridized Algorithms for a Customer Information Profiling System

Abstract

Accurate customer information profiling remains a critical challenge for banks and businesses due to the massive volume of customer data and the need for timely, targeted service delivery. The conventional Back Propagation Neural Network (BPNN) algorithm suffers from high computational time, limiting its efficiency in large-scale data analysis. To address this limitation, this study developed a hybridized MapReduce Back Propagation Neural Network (MRBPNN) technique for customer information profiling. A dataset comprising 2,000 customers’ demographic and socioeconomic records, each with 35 attributes and 5 status updates, was obtained from selected banks in Edo State. The data were divided into training and testing sets and evaluated using a ten-fold cross-validation approach. The hybrid MRBPNN model was implemented in MATLAB R2023a and compared with the conventional BPNN algorithm based on false positive rate (FPR), precision, recall, F-measure, accuracy, and computational time. Experimental results revealed that MRBPNN achieved superior performance with an FPR of 5.83%, accuracy of 94.59%, precision of 99.31%, recall of 94.64%, F-measure of 96.92%, and computational time of 13.23 seconds. In contrast, BPNN recorded an FPR of 15.61%, accuracy of 84.97%, precision of 97.96%, recall of 85.04%, F-measure of 91.05%, and computational time of 26.69 seconds. The findings demonstrate that the developed MRBPNN significantly reduces computational time and enhances classification accuracy compared to BPNN, making it an efficient and scalable technique for customer information profiling in financial institutions.