Hybrid Clustering and Optimized DBN for Data-Driven Visual Merchandising Insights within the Textile Sector

The main goal of this study is to improve the effectiveness of customer prediction in order to benefit businesses.To achieve this objective, the research focuses on addressing two main challenges.Firstly, it solves the issue of clustering large business datasets.For this purpose, it introduces a novel approach called hybrid K-means++ with EM clustering.This method aims to better organize the data points into distinct clusters, enhancing the understanding of customer behaviour patterns.Secondly, the research aims to enhance the efficiency of training models for customer prediction, particularly when dealing with big data.It achieves this by employing Dynamic Adjustable Momentum Backpropagation (DAMB) within Deep Belief Networks (DBN).By utilizing DAMB, the research optimizes the training process, resulting in more accurate prediction models.In addition to these primary objectives, the study also contributes by developing and utilizing two custom datasets: the South India Visual Merchandising Survey (SIVMS) and the Global Visual Merchandising and Customer Behaviour Survey (GVMCBS).These datasets provide a broader perspective on customer behaviour in visual merchandising settings, including diverse demographics and geographical regions.To assess the effectiveness of the proposed methods, the research conducts thorough analyses in three key areas: clustering efficiency, training efficiency, and computational efficiency.The proposed method demonstrates superior clustering results compared to traditional K-means and K-means++ algorithms when applied to the SIVMS and GVMCBS datasets.Specifically, it resulted in less overlap between customer groups and greater separation between the centers of those groups.Additionally, the proposed method offers slightly improved testing efficiency.In comparison to standard Backpropagation (BP) models and DBN variants (Deep Belief Networks-Backpropagation (DBN-BP), Deep Belief Networks-Backpropagation-Adaptive Learning Rates (DBN-BP-ALR), Deep Belief Networks-Resilient Backpropagation (DBN-RBP), Deep Belief Networks-Levenberg-Marquardt Algorithm (DBN-LMA), and Deep Belief Networks-Conjugate Gradient (DBN-CG)), the proposed Deep Belief Networks-Dynamic Adjustable Momentum Backpropagation (DBN-DAMB) model yields the most favourable results.The proposed DBN-DAMB model also has the lowest computational demands, with 1.8 million parameters and 2.5 x 10 9 ) Floating-Point Operations Per Second (FLOPs).These results show that the system can accurately predict customer behavior and is efficient in terms of computation.