The Internet of Things (IoT) consists of interconnected devices that continuously exchange data, making security a critical concern. However, a number of issues pertaining to IoT security and privacy have emerged as a result of its broad use. For IoT-enabled services to be dependable, secure, and profitable, real-time intrusion detection on IoT devices is essential. Current Intrusion Detection Systems (IDSs) frequently face challenges such as a high False Alarm Rate (FAR), mean squared error, and reduced intrusion detection reliability and accuracy. To address intrusion detection challenges in IoT environments, this work proposes the Shrike bird Optimization and concrete Autoencoder-based deep Learning framework for IOT intrusion detection (SOAL-IOT), as illustrated in the diagram. The process begins with IoT network traffic collection, followed by data pre-processing, which integrates data cleaning and normalization to improve high-quality data. Then, Feature Extraction (FE) is performed using a Concrete Autoencoder (CAE) to learn compact and meaningful feature representations. The extracted features are then refined through Shrike Bird Optimization (SBO) for effective feature selection, which reduces redundancy and computational complexity. Finally, the selected features are classified using a Neural Controlled Differential Equation (NCDE)-based model to accurately distinguish between normal and malicious traffic. Experimental findings on the BoT-IoT datasets demonstrate that the proposed model improves overall accuracy by 4.93%, 6.26% and 4.23% over SPOHDL-ID, CST- AFNet, and HIDSIoMT on the BoT-IoT Dataset.