Optimizing identity and access management through 1D-SCNN-based anomaly detection

Identity and Access Management (IAM) systems are critical in the ever-evolving digital landscape as legacy security methods fall short against modern cyber threats. This study proposes a fast and accurate anomaly detection method named 1D-Separable Convolutional Neural Networks (1D-SCNN), effectively detecting abnormal identity access or credential abuse. This method utilizes deep learning to analyze user activity and access habits from a one-dimensional structure, leveraging the benefits of 1D-SCNN, such as lower computational cost and higher model efficiency. The proposed model employs a 1D-SCNN architecture customized for efficient anomaly detection in IAM systems. It uses separable convolutions to handle one-dimensional input data, reducing the number of parameters and required computation. The architecture includes layers such as Leaky ReLU and ELU for activation, MaxPooling for down-sampling features, Dropout for regulating overfitting, and a Flatten layer for classification. This configuration allows the model to learn from historical user engagement data and identify anomalous behavior patterns, which are strong indicators of security threats. The study results highlight the value of advanced deep learning techniques in cybersecurity and provide a roadmap for integrating 1D-SCNN within IAM systems to enhance security in digital environments. Finally, in experiments on an extensive data set, the proposed model outperformed by achieving an impressive accuracy of 96%.