CICIDS2017 dataset used in this research paper comes from the Canadian Institute for Cybersecurity and comprises of labelled and tagged benign and botnet traffic in PCAP files. We also used three NSL KDD IoT datasets - kddubei (Ghernaouti, 2012), kddprediction (2012), kddaccessb (2012) obtained from the Australian Defence Science and Technology Group. In this paper, we perform a systematic and comparative performance evaluation of commonly used machine learning and deep learning algorithms to detect network intrusions. We evaluated the performance of these classification algorithms against four different datasets. The main focus being the use of these algorithms specifically in NIDSs, our comparative analysis will also focus on how these algorithms fare under various NIDS specific issues such as imbalanced data, scoring, etc. In recent times, there has been a growing trend of using machine learning (ML) and deep learning (DL) algorithms. Although much research has been conducted on IDSs, especially using NSL-KDD datasets, that propose various methods to detect intrusions and achieve high detection rates, the use of ML and DL in NIDSs is still relatively new, and especially there are only a scarce number of studies that perform a comprehensive and systematic comparison of the performances of various machine learning algorithms and deep learning models. NIDSs are one of the most prominent components of any network's security infrastructure. With the rise in cyber-crime, NIDS usage and deployment have become an essential need. A number of approaches have been devised to detect malicious traffic across the networks, and all of them employ different techniques to either store the signatures of known intrusions and then use these signatures to match the incoming traffic packets through machine learning algorithms.