Customer Segmentation and Outlier Mitigation for Electronic Toll Collection System (ETC): A Case Study of Burapha Withi Expressway, Thailand

The development of Machine Learning and Data Science, combined with the rapid growth of transaction activities in Electronic Toll Collection (ETC) systems highlights the customer segmentation for meeting diverse customer needs as an effective marketing strategy. By employing K-Means clustering and the Length-Recency-Frequency-Monetary (LRFM) model, this study aims to uncover actionable insights into customer behavior for revealing valuable business insights and strategic opportunities. The Elbow Method, Silhouette Score, and Davies-Bouldin Index were utilized to validate the clusters. While handling numerous outliers in K-means clustering can be a bit more challenging, clustering the outliers can interpret hidden business insights or a story seeking. Therefore, it’s crucial to ensure this method aligns with business goals rather than just removing as a data convenience. Using the ETC toll dataset from Burapha Withi Expressway, Thailand (28,309,012 transactions and 1,249,383 customers), the segmentation identified four clusters without outliers (0: Nurture, 1: Reward, 2: Re-engage and 3: Retention) and three clusters with outliers (-3: Delight, -2: Upsell, and -1: Pamper). The delight segment has the most recent recency value, the highest frequency, and the most significant monetary value. The results of this customer segmentation became a consideration for the highway organization in making marketing strategies that are right on target and the framework achieved the Silhouette Score of 0.410 and the Davies-Bouldin Index of 0.830.