A comprehensive ATM security framework for detecting abnormal human activity via granger causality-inspired graph neural network optimized with eagle-strategy supply-demand optimization

被引：0

作者：

Kshirsagar, Aniruddha Prakash ^{[1
]}

Azath, H. ^{[2
]}

机构：

[1] VIT Bhopal Univ, Sch Comp Sci & Engn, Sehore, Madhya Pradesh, India

[2] VIT Bhopal Univ, Sch Comp Sci & Engn, Sehore, Madhya Pradesh, India

来源：

EXPERT SYSTEMS WITH APPLICATIONS | 2025年 / 272卷

关键词：

Abnormal Human Activity; ATM Security; Eagle-Strategy Supply-Dem & Optimization; Granger Causality-Inspired Graph Neural; Network; Reverse Lognormal Kalman Filtering;

D O I：

10.1016/j.eswa.2025.126731

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

The daily increase of criminal activity has made real-time human activity detection crucial for the protection & surveillance of public spaces, including bank-automated teller machines (ATM). To overcome the difficulty of online identification of anomalous activity in bank automated teller machines. A Comprehensive ATM Security Framework for Detecting Abnormal Human Activity via Granger Causality-Inspired Graph Neural Network optimized with Eagle-Strategy Supply-Dem& Optimization (ATM-DAHA-GCIGNN-ESSDO) is proposed in this manuscript. Initially, the input videos are gathered from DCSASS Dataset & UCF Crime Dataset. Then, the video is pre-processed by using Reverse Lognormal Kalman Filtering (RLKF) for cleaning noisy data. Granger CausalityInspired Graph Neural Network (GCIGNN) is employed for detect abnormal human activities in ATM machine. Abuse, Arrest, Arson, Assault, Burglary, Explosion, Fighting, Road Accidents, Robbery, Shooting, Shoplifting, Stealing, V&alism for DCSASS Dataset & Abuse, Arrest, Assault, Arson, Burglary, Explosion, Fighting, Normal Videos, Road Accidents, Shoplifting, Shooting, Robbery, Stealing, V&alism for UCF Crime Dataset. The EagleStrategy Supply-Dem& Optimization (ESSDO) is implemented to enhance the parameters of GCIGNN. The proposed method is implemented & the efficiency is estimated using some performance metrics, like Accuracy, Recall, F1-score, precision, False Discovery Rate & Computational time. The performance of the ATM-DAHAGCIGNN-ESSDO approach attains 24.39%, 35.71%, & 25.55% higher Accuracy; 22.15%, 24.21%, & 43.52% higher Recall. The proposed ATM-DAHA-GCIGNN-ESSDO framework outperforms the existing approaches for identifying aberrant human activity in ATM & criminal situations. Finally, the proposed approach demonstrates its potential as a reliable solution for real-time security & surveillance applications.

引用

页数：13