Towards Detecting And Classifying Network Intrusion In The Internet Of Things

Putro, Iwan Handoyo (2026) Towards Detecting And Classifying Network Intrusion In The Internet Of Things. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7025231003-Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

The rapid expansion of Internet of Things (IoT) networks has heightened critical security vulnerabilities, yet existing machine learning-based Intrusion Detection Systems (IDS) remain hampered by sub-optimal classification accuracy, limited benchmark evaluations, and fragmented pipelines. Addressing these gaps, this thesis introduces a holistic, end-to-end hybrid machine learning framework designed for high-accuracy anomaly detection and threat classification in IoT environments. Unlike conventional IDS frameworks that treat data preprocessing and model training as isolated phases, our architecture tightly bridges data-level feature optimizations with model-level algorithmic enhancements into a unified, interdependent pipeline. This co-designed approach integrates domain-tailored feature engineering, strategic feature selection, and class imbalance mitigation into a single pipeline. By systematically filtering redundant IoT traffic while amplifying low-frequency attack signatures, the framework resolves structural benchmark limitations including high dimensionality, noisy traffic, and severe class disparity without escalating computational overhead. Evaluated across two benchmark datasets (UNSW-NB15 and RT-IoT 2022), the proposed model significantly outperforms baseline approaches, achieving an overall Accuracy of 98.90%, an F1- score of 99.39%, and a sharp reduction in False Positive Rate (FPR) from 63.98% down to 11.79%. These results demonstrate a scalable, highly precise solution optimized for the constraints of modern IoT infrastructure.
========================================================================================================================================
Pesatnya pertumbuhan jaringan Internet of Things (IoT) telah memicu peningkatan serangan siber yang mengakibatkan kerugian signifikan. Machine learning telah banyak diterapkan untuk meningkatkan kinerja Intrusion Detection Systems (IDS), namun penerapannya masih terhambat oleh akurasi klasifikasi yang sub-optimal, keterbatasan evaluasi dataset, dan alur kerja (pipeline) yang terfragmentasi. Untuk mengatasi permasalahan tersebut, tesis ini memperkenalkan sebuah kerangka kerja hybrid machine learning yang holistik dan dirancang untuk mendeteksi anomali dan klasifikasi serangan siber pada lingkungan IoT. Berbeda dengan kerangka kerja IDS konvensional yang memperlakukan pra-pemrosesan data dan pelatihan model sebagai fase terpisah, arsitektur yang diusulkan menghubungkan secara erat optimasi fitur tingkat data dengan peningkatan algoritmik tingkat model. Pendekatan ini mengintegrasikan rekayasa fitur berbasis domain, seleksi fitur, dan mitigasi ketidakseimbangan kelas ke dalam satu pipeline tunggal. Kerangka kerja ini mampu menyelesaikan keterbatasan struktural pada benchmark, seperti dimensionalitas tinggi, traffic data berderau (noisy traffic), dan ketimpangan kelas, tanpa meningkatkan beban komputasi. Dievaluasi menggunakan dua dataset benchmark (UNSW-NB15 dan RT-IoT 2022), kerangka kerja ini secara signifikan mengungguli model pembanding, dengan mencapai akurasi keseluruhan sebesar 98,90%, F1-score sebesar 99,39%, serta penurunan tajam pada False Positive Rate (FPR) dari 63,98% menjadi 11,79%. Hasil penelitian ini dapat mengatasi kendala keamanan pada infrastruktur IoT modern.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Machine Learning. Internet of Things, Intrusion Detection System
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.5956 Quality of service. Reliability Including network performance
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55001-(S3) PhD Thesis (Comp Science)
Depositing User:	Iwan Handoyo Putro
Date Deposited:	02 Aug 2026 14:53
Last Modified:	02 Aug 2026 14:53
URI:	http://repository.its.ac.id/id/eprint/141986

Actions (login required)

View Item