<![CDATA[Trader activities at trading firms exhibit heterogeneous characteristics, necessitating an analytical approach to identify segmentation patterns and trader activity status. This study aims to analyze GKInvest trader segmentation using the K-Means algorithm and to build a Random Forest model to predict trader activity status based on historical activity characteristics. The dataset comprises trader data with key variables including 30-day transaction volume, 30-day login count, 30-day transaction count, days since last activity (recency), account tenure, number of active days, and the number of instruments traded. K-Means was employed as an unsupervised learning method to establish segments, while Random Forest was used independently as a supervised learning method to classify active versus inactive status. The K-Means analysis yielded three segments: Passive Traders (Churn Risk) comprising 1,135 traders (56.7%), Highly Active Traders (VIP) comprising 437 traders (21.9%), and Regular Active Traders comprising 428 traders (21.4%). The Random Forest model achieved an accuracy of 83%; performance for the "Active" class reached a precision of 84%, recall of 93%, and F1-score of 88% (with a weighted average F1-score of 82%). Feature importance analysis revealed that 30-day transaction volume made the largest relative contribution (~0.218), followed by 30-day login count (~0.184) and 30-day transaction count (~0.183). The findings demonstrate that K-Means and Random Forest can independently provide complementary perspectives for understanding the activity characteristics and status of GKInvest traders, serving as a foundation for developing data-driven trader activity monitoring systems.]]>
