erenonal/K-means_customer_segmentation

Implements a complete preprocessing pipeline including outlier detection via IQR bounding, StandardScaler normalization, and the Elbow method for optimal cluster selection, operating on ~9000 credit card holders across 18 behavioral variables. The approach applies cosine similarity before PCA to preserve vector relationships, then reduces dimensionality for cluster visualization and interpretation across six distinct customer segments (high spenders, cash advance users, installment buyers, etc.). Built with scikit-learn, pandas, and matplotlib, enabling direct actionable marketing strategies like credit limit adjustments and targeted product offerings per segment.

No commits in the last 6 months.