The rapid development in computer network and internet have resulted in increased corresponding data and network attacks. Many novels and improvement technologies have been developed in the literature to handle security issues in the network and accurately detect intrusions. Intrusion Detection System (IDS) is one of the technological tools that aim to ensure confidentiality, integrity and availability of the system by preventing possible intrusions. Despite its effectiveness in reducing network intrusion, the IDS still face some challenges in improving detection accuracy and False Error Rate (FER). Recently, researchers have shifted toward the use of Machine Learning (ML) and Deep Learning (DL) methods to enhance the performance of IDS by increasing accuracy rate and reducing (FER). The improvement in the performance of IDS system depends on the detection method used, benchmark dataset and IDS environment. This paper aims to provide a comprehensive survey for the most recent articles that have been published between 2018 and 2021 which used ML and DL methods to improve IDS accuracy. The selected papers are analysed to cover an overview about the following aspects: (1) IDS concepts and taxonomy. (2) ML and DL methods along with strengths and weaknesses for each one. (3) Benchmark datasets of IDS. (4) Comprehensive review about the most recent articles in this domain with the advancement provided by each work in terms of methodology and dataset used with highlighting the strengths and weaknesses of each work. Finally, the challenges and the future scope for the research in IDS based ML and DL are provided.