Internet users are rapidly growing all over the world. Hence, cybersecurity is a significant concern in protecting electronic devices from intruders worldwide. The attacker attempts novel denial-of-service (DoS) attacks over the network device to deny or disturb the services. The digital devices are not secure themselves. Therefore, it is essential to intend an efficient intrusion detection system to detect the novel types of attacks intelligently to protect the devices. The present study suggests feature selection using information gain (IG) by performing threshold values to produce subsets. The analysis introduces the attack detection system to effectively detect ARP poisoning, scans, DoS, and exploitation-based attacks. The system is tested on the UKM-IDS20 dataset with a random forest classifier for multilevel-based and binary classification. The experiments on the UKM-IDS20 dataset show the proposed method specifies an correctness of 99.9845% with random forest on 17 relevant features.