Identifying the Latitude and Longitude of ATMs in ATM Networks

The geographical positioning of Automated Teller Machines (ATMs) is a pivotal data point that significantly aids in the analytical process and decision-making for a multitude of critical banking and economic determinations. Given the constraints imposed by the insular viewpoint prevalent in the nation’s banking ecosystem, maintaining a consolidated perspective of all ATMs’ geographical locations is not feasible. In this study, we utilized the ATM Location Prediction (ATMLP) algorithm to determine these machines’ geographical coordinates. This data is indispensable and plays a cardinal role in the implementation of a multitude of artificial intelligence algorithms.

The ATMLP algorithm comprises three primary stages. The first stage involves constructing a bipartite user-location graph. The relationship between users is derived from transactional interactions, while the relationship between geographical locations is established using devices with known locations. The second stage involves the computation of two crucial indices: spatial similarity and neighborhood similarity, within the ATM network using the bipartite graph. This stage also includes a time-space distance finding module, which has two steps in its procedure: finding co-located ATMs and then clustering them. Distance-based features are assigned to edges because they reflect the similarity level between the pair of ATMs, nodes connected by edges. The third stage of the algorithm fine-tunes the results for better accuracy. In this process, low-confidence edges are filtered out by leveraging similarity metrics from the previous stage and cosine similarity between pairs of ATMs. In the end, the algorithm reports the geographical latitude and longitude for each ATM, plus the probability score indicating how correct it is.

By leveraging 2100 ATM locations (a portion of the data available in Datis Arian Qeshm Company) and examining 562609790 transactions in four months from the start of April 2022 to the end of July 2022, we identified the location of 4000 existing ATMs across the country belonging to 12 banks. The results obtained indicate a high credibility of the algorithm (80.95%).

In this study, we applied a developed method in banking to predict edges in location-based social networks, and using it, we accurately estimated the geographical coordinates - latitude and longitude - of ATMs on a national scale. Location-based social networks, due to data integration at multiple levels, enable problem-solving that was previously impossible. The use of these methods has less processing cost and higher speed due to the use of algorithms and graph-based databases, and they provide more accurate results. This study has significant implications for future research in banking technology, particularly about location prediction for ATMs.