Multi-Level Access Control System in Automated Teller Machines

E-commerce theft involves using lost/stolen debit/credit cards, forging checks, misleading accounting practices, etc. Due to carelessness of cardholders and criminality activities of fraudsters, the personal identification number (PIN) and using account level based fraud detection techniques methods are inadequate to cub the activities of fraudsters. In recent times, researchers have made efforts of improving cyber-security by employing biometrics traits based security system for authentication. This paper proposed a multi-level fraud detection system in automated teller machine (ATM) operations. The system included PIN level, account-level and biometric level. Acquired RealScan-F scanner was used to capture liveness fingers. Transactional data were generated for each individual fingerprint with unique PIN. The results of the simulation showed that (i) the classification at account level only yielded averages 84.3% precision, 94.5% accuracy and 5.25% false alarm rate; (ii) matching at biometric level using liveness fingerprints samples yielded 0% APCER , 0% NPCER and 100% accuracy better than using fingerprints samples that produced 4.25% APCER , 2.33% NPCER and 93.42% accuracy; (iii) combining the three levels with the condition that all the levels must be positive produced 87.5% precision,84.9% accuracy and 2.65% false alarm rate; (iv) while the classification using voting technique yielded 99.15% precision, 97.35% accuracy and 0.47% false alarm.