Komparasi Teknik Feature Selection Dalam Klasifikasi Serangan IoT Menggunakan Algoritma Decision Tree

Abstract

Presence of Internet of Things (IoT) has revolutionized how we interact with the world on our daily life by enabling various devices to connect the internet and transmit data. However, the increasingly widespread use of IoT technology also brings serious threats to cyber security and increases the number of IoT attacks. The need for robust classification models is becoming increasingly clear to anticipate these problems. This research focuses on developing an IoT attack classification model by comparing feature selection techniques that utilize data from the CIC IoT Dataset 2023. This research faces challenges such as data imbalance and the complexity of handling various features. To overcome these challenges, this research uses random undersampling techniques to balance the data and utilizes various feature selection methods, including filter based, wrapper based, and embedded based. Apart from that, this research also tries to use a decision tree algorithm. The findings reveal that the application of wrapper based techniques as feature selection together with a decision tree algorithm produces the highest accuracy of 87.32% in classifying IoT attack types. This emphasizes that the use of techniques and algorithms that are still rarely used can provide fairly good accuracy results.