Big Data Analytics, Greedy Approach, and Clustering Algorithms for Real-Time Cash Management of Automated Teller Machines

Abstract

Automated Teller Machines (ATMs) often lack the required funds or become malfunctioned, which affects the customer experience and the reputation of the bank. Banks try to quickly resolve the problem through cash-in-transit companies that handle the operations of ATM refilling and maintenance. However, one of the largest dilemmas is to determine the order of visiting the ATMs as well as to balance the workload among the workforces during the day. In addition, there is a need to handle real-time and urgent requests during the day. This problem was modelled as a realtime multiple Travelling Salesmen Problem (mTSP). New constrains including traffic data, ATM priorities, and safety measurements were considered. We used big data analytics to extract useful features related to the customer withdrawal trends and active locations from real data provided by a Bahraini bank. To solve this NP-hard problem, we proposed a brute force method that generates optimal routes for limited-sized problem instances, up to 35 ATMs. Moreover, a greedy technique was proposed to solve large-sized instances considering one salesman. The obtained TSP route is then cut into clusters using unsupervised machine learning models. A modified version of k-Means has been applied with constrains to control the size of each cluster.