Customer Behavior Clustering Based on Balance History Using Dynamic Time Warping Distance

Abstract

Customer clustering, the division of customers into different groups, is a classical problem. It is especially important in banking as it serves multiple purposes in marketing, risk management, etc. Therefore, it has attracted the use of many modern machine learning models and techniques. But currently, most of them are only making use of "static" customer information. This paper proposes a new approach for customer clustering in banking based on the customers' balance history. Basic Dynamic Time Warping (DTW) distance and Soft-DTW (SDTW) distance, an advanced form, are used to measure the difference between customers. To which, the two most popular strategies in time series clustering strategies, partitional and hierarchical, are applied which. In additional, some statistical features are given to prove the effectiveness of the proposed method.

Hybrid Multi-Infeed Receiver Line Longitudinal Protection Scheme Based on Voltage Waveform Comprehensive Distance Similarity

“…Euclidean distance, also known as Euclidean metric, is commonly used to measure the true distance between two points in a space, reflecting the overall distribution characteristics of waveform similarity [21]. The instantaneous values of measured voltages at the protection installation points M and N are extracted.…”