Machine Learning — Unsupervised Clustering
This project applies unsupervised machine learning to the Beverage Sales dataset, containing 8,999,910 transaction rows and 11 original features. K-Means clustering is used to discover purchasing patterns from customer type, unit price, quantity, discount, and total transaction value.
The dataset contains beverage sales transactions with numerical and categorical variables. After preprocessing, the clustering model uses Customer_Type, Unit_Price, Quantity, Discount, and Total_Price.
Dataset Loading → Exploratory Data Analysis → Feature Selection → Label Encoding → Standardization → K-Means Clustering → Elbow & Silhouette Evaluation → PCA → Cluster Interpretation
The clustering model was configured with 3 clusters using K-Means.
The sampled Silhouette Score reached approximately 0.712 before PCA-based optimization.
After PCA retained 95% of the variance, the sampled Silhouette Score reached 0.7239.
Cluster 0 represents lower-value purchases with an average unit price of 4.47, average quantity of 7.99, and average total price of 35.17.
Cluster 1 represents high-volume purchases with an average quantity of 49.65 and average total price of 135.18.
Cluster 2 represents premium, high-value transactions with an average unit price of 72.70 and average total price of 2,425.09.
