AI-Driven Anomaly Detection for Secure Identity and Access Management in Cloud Platform

In recent years the use of the cloud computing solutions raised the consciousness of the effective protection of the identity and access management systems. The mainstream anomaly detection technology includes traditional machine learning approach and basic artificial neural network which have limitations in efficiency and accuracy, which results in significant weakness in detecting complex attack. This research introduces a novel technique of using Long Short-Term Memory Autoencoder (LSTM-AE) that has lesser disadvantages of traditional machine learning models in capturing intricate details from the data appropriately. As the suggested methodology it is formed by gathering and preprocessing a large dataset for training LSTM-AE with a comparison to GRN, 1D CNNs, TCNs, and AE. The accuracy, precision, recall, and F1 result showed that LSTM-AE achieved result in 0.997 accuracy and 0.998 F1, meaning that LSTM-AE has the highest anomaly detection capability when compared to the other models. On this basis, the assumptions made in the proposed model while strengthening the possibilities of using IAM systems for the detection of identity threats and other security incidents also make a notable contribution to the enhancement of effective methods for managing security threats in cloud computing. Subsequent works will advance into realtime applications and enhancements of this strategy and, reinforcing its potential as a critical tool in safeguarding sensitive information against emerging threats.