The use of networks in modern life makes cyber security an essential field for research. An essential cyber security tool is an intrusion detection system (IDS), which keeps an eye on the state of the hard- ware and software running on the network. IDSs on the market today still struggle to detect unknown attacks, reduce false alarm rates, and improve detection accuracy despite years of research. Numerous studies have been conducted to develop IDSs that employ machine learning methods to tackle the previously mentioned problems. Automatically recognizing the salient features that differentiate normal from anomalous data is a highly accurate task for machine learning techniques. Furthermore, the great generalizability of machine learning techniques enables them to identify threats that are yet unknown. Deep learning is a highly performing subfield of machine learning and is a popular research topic. This study proposes an IDS taxonomy that bases its classification and compilation of deep learning and machine learning-based IDS literature on data objects as the main dimension. Cyber security researchers should, in our opinion, use this type of taxonomy structure. The taxonomy and concept of IDSs are introduced at the outset of the survey. Next, machine learning techniques commonly used in IDSs are presented, along with measurements and benchmark datasets. Then, we show how to tackle significant IDS problems with machine learning and deep learning techniques, using the proposed taxonomic system as a baseline and in conjunction with sample literature. Lastly, challenges and future directions are investigated through an assessment of representative recent research.