Keylogging malware is a grave risk to user credentials and data integrity in the constantly evolving discipline of cybersecurity. In an effort to conquer this obstacle, our study optimizes early keylogging detection by creating a Neuro-fuzzy prediction model based on keystroke dynamics. The model was trained on a dataset of more than 500,000 keystroke samples from real keyloggers and simulated users by combining adaptive neural networks and fuzzy logic inference. The tailored Neuro-fuzzy model clearly reduced false positives, increasing accuracy to 99.62% and precision to 66.67, compared to the initial neural networks’ 99.1% detection accuracy. A 0.378 MSE is a performance indicator that highlights the model’s resilience. By identifying unusual keystroke patterns prior to leaking data, our technology offers an early defense against keylogging, which is a major improvement over conventional defensive defense.
Keywords: keylogging, keylogging threats, neuro-fuzzy, neuro-fuzzy model
