Internet of Things (IoT) plays a significant part in the functioning of complex infrastructure. Due to the number of complex and noteworthy data replaced across the comprehensive internet, intrusion detection system (IDS) is a stimulating operation in numerous vast IoT networks. Therefore, this research proposes the feature selection approach of the Golden Sine-based Spotted Hyena Optimization algorithm (GS-SHO) approach is proposed for the IDS in IoT. The benchmark IDS datasets like CICIDS2017 and NSL-KDD are utilized for estimating the efficacy of the proposed method. Then, the min–max normalization technique is utilized for preprocessing the data. Eventually, the bidirectional long short-term memory (BiLSTM) is utilized for the classification of IDS into two types such as normal and attack. The proposed GS-SHO approach attains better accuracy of 99.56% and 98.96% on CICIDS2017 and NSL-KDD datasets when compared to the existing methods like Optimized Common Feature Selection as well as Deep Autoencoder (OCFSDA), and Improved Binary Golden Jackal Optimization (IBGJO) approach.