As the number of Internet of Things (IoT) devices grows, they are prone to a large number of security threats, as reported repeatedly in recent years. Several research works have employed machine learning-based techniques to develop intrusion detection systems (IDS) that distinguish abnormal behavior of IoT devices from a normal network activity. Due to the statistical variations in data over period of time, single classification approaches are not efficient enough to tackle the abnormalities in traffic. In such cases, when the data volumes are higher and data is time variant, ensemble learning based stacking approaches can be adopted as a better alternative to improve prediction quality. In this project, we aim to design novel feature extraction and selection algorithms for the IDS system. Our objective is to develop a stacking-based intrusion detection system using ensemble learning techniques that efficiently combines the prediction of individual algorithms to generate a combined optimized prediction. We will test the effectiveness of the proposed solution on the multiple state-of-art datasets, e.g., KDD99, NSL_KDD, and BoT-IoT. The goal is to improve the precision, accuracy, and prediction quality of intrusion detection systems.