<p>Internet of Things (IoT) is a networked system that effectively connects the computer as well as other devices to enable seamless communication and data exchange through telecommunication technologies. However, security threats remains challenging in the IoT environments. Developing an effective anomaly detection method is required to recognize sophisticated sample from noisy data, detect irregularities in the absence of sufficient labelled data and identify dynamic anomaly behaviors. In this research work, a novel deep learning-driven anomaly detection method is employed to provide security in an IoT environment. Initially, the required IoT time series data is collected from standard databases. It further undergoes through cleaning and scaling process to prevent from invalid or irrelevant entries, thereby enhancing data quality. From this high-quality data, the PCA, statistical and t-SNE features are separated and it is fused to capture relevant information from the IoT time series data effectively. Finally, the fused features are input into an Autoencoder-based Adaptive Bidirectional Long Short Term Memory with Attention Mechanism (AABiLSTM-AM) for detecting irregular operative conditions. Hyperparameter tuning is performed to further improve the detection accuracy using the Advanced Gold Rush Optimization Algorithm (AGROA). The empirical validation of the developed approach is validated over conventional anomaly detection techniques to ensure effectiveness based on diverse measures. In the estimation phase, the proposed model has attained 98.67% accuracy and 98.41% precision to demonstrate an efficient anomaly detection performance.</p>