Customer Behavior Analytics Using Manifold Learning and Ensemble Clustering

Customer segmentation plays a crucial role in
analysing purchasing patterns of customers in large scale
retail industry. However, transaction data are usually highdimensional
and noisy which limits the effectiveness of
conventional clustering techniques. This paper presents an
integrated framework that combines manifold learning and
ensemble clustering to make improved and robust customer
behavior analysis. Initially RFM and other insightful
purchase behavior features are extracted from the
transaction record of online retail dataset. UMAP is then
applied to reduce the dimension and also preserves the
intrinsic data structure. HDBSCAN is employed to identify
high density clusters and detect noise. To enhance quality of
cluster, MiniBatch K-Means is applied in core clusters and a
GMM assigns noisy customers based on probabilistic
similarity. The experimental results on an online retail dataset
demonstrate the robust and interpretable customer segments.
The proposed framework improves clustering stability and
also supports effective data-driven customer analytics and
targeted marketing strategies.