1. Introduction
Distributed Denial of Service attack detection is the integral content in network security, and it is widely applied in securing data from intruders. DDoS is a cybercrime attack that causes users to access services by flooding many requests. The cybersecurity industry reported that DDoS attacks are the most prevalent threats in recent research. Since these attacks disguise the hacker's identity using third-party tools and components, thereby allowing the attacker to attack the server as a normal packet. Also, these attacks are passive as they do not steal information but intercept packets that reach the server. The greater the number of connected devices, the more the chance of botnet formation, resulting in DDoS attacks. Due to the massive growth of data in the network, the need for a feature selection algorithm is undoubtedly increasing. The reason feature selection has enjoyed increased attention is a lot of irrelevant and redundant data. Nowadays, varieties of feature selection methods have been proposed by researchers, but they deal with less number of features. The dataset CICDDoS2019 evaluation dataset used in our paper is enormous, it contains many irrelevant and redundant features. Removing these extra features should be done to avoid loss of information and learning performance degradation. Feature selection on conventional data has been challenging work. This paper focuses on selecting the top ten features that significantly affect detecting DDoS attacks. Further, we have analyzed the classification accuracy achieved when Chi-Square, Extra Tree, ANOVA and Mutual Information are Integrated with random forest and decision tree. The paper is organized as follows: Section 2 briefly introduces the related work according to the frameworks. Section 3 presents the proposed FSMDAD model, dataset description, hardware specifications, and machine learning algorithms. Section 4 describes the results and discussions. Conclusion is presented in Section 5.