The advance of internet usage in Indonesia gives positive impact on the development of e-commerce in Indonesia where 63.5% of internet users have made online transactions. Along with ecommerce B2C growth in Indonesia, firm needs for an effective promotional strategy to understand the preferences and potential purchases for each consumer to increase return on investment (ROI). This empirical study investigated purchase decision of ecommerce users using Web Usage Mining framework. The high combination of purchasing product categories by users of ecommerce website required a multi-label classification technique that could classify those pair of purchase decision. Label Powerset method with Support Vector Machine (SVM) algorithm was applied to classify e-commerce users purchase decisions using general and detailed information features. Feature selection using Information retained 60 from 90 features. The proposed feature selection with Information Gain and parameter selection using Grid Search proved that they had an ability to enhance performance to classify purchase decision of e-commerce user. Radial Basis Function (RBF) as the most effective kernel presented an accuracy of 75.6%, with slightly difference of 2.2% with classification without using feature selection.