IKHWAN, Syariful and Purwanto, Purwanto and Rochim, Adian Fatchur (2026) PENINGKATAN KINERJA DEEP FEEDFORWARD NEURAL NETWORK UNTUK DETEKSI INTRUSI PADA JARINGAN KAMERA IP MENGGUNAKAN METODE SPADEH SEBAGAI PENDEKATAN ENSEMBLE FEATURE SELECTION. Doctoral thesis, UNIVERSITAS DIPONEGORO.

	Text cover aja.pdf Download (37kB)
	Text .Cover-Daftar Istilah.pdf Restricted to Repository staff only Download (884kB)
	Text BAB I.pdf Download (111kB)
	Text BAB II.pdf Download (813kB)
	Text BAB III.pdf Restricted to Repository staff only Download (2MB)
	Text BAB IV.pdf Restricted to Repository staff only Download (2MB)
	Text BAB V.pdf Restricted to Repository staff only Download (30kB)
	Text Daftar Pustaka.pdf Download (212kB)

Abstract

Perkembangan pesat teknologi Internet of Things (IoT), khususnya pada perangkat multimedia seperti kamera IP, telah meningkatkan potensi ancaman keamanan siber yang semakin kompleks. Sistem deteksi intrusi (Intrusion Detection System/IDS) berbasis machine learning telah banyak digunakan untuk mengidentifikasi serangan jaringan, namun masih menghadapi berbagai kendala, seperti akurasi yang belum optimal, tingginya false alarm rate, serta kesulitan dalam menangani data berdimensi tinggi. Selain itu, karakteristik lalu lintas data pada jaringan kamera IP yang dinamis menuntut pendekatan deteksi yang lebih adaptif dan efisien. Oleh karena itu, diperlukan metode yang mampu meningkatkan kualitas representasi fitur serta performa model deteksi intrusi secara menyeluruh. Penelitian ini mengusulkan pendekatan deteksi intrusi berbasis Deep Feedforward Neural Network (DFNN) yang dioptimalkan menggunakan metode SPADEH (Selective Predictive Attack Detection using Ensemble Hybridization). SPADEH merupakan metode ensemble feature selection yang mengintegrasikan berbagai perspektif hubungan antar fitur melalui mekanisme fusion untuk menghasilkan subset fitur yang lebih representatif dengan mempertimbangkan hubungan linear, monotonik, dan non-linear antar fitur. Model yang diusulkan dievaluasi menggunakan dataset publik UNSW-NB15 serta dataset real-world yang diperoleh dari lalu lintas jaringan kamera IP. Hasil penelitian menunjukkan bahwa pendekatan yang diusulkan mampu meningkatkan performa deteksi intrusi secara signifikan dibandingkan metode baseline. Kombinasi seleksi fitur yang digunakan terbukti efektif dalam mengurangi dimensi data sekaligus meningkatkan akurasi dan stabilitas model. Model DFNN yang dioptimalkan juga mampu menurunkan false alarm rate serta menunjukkan kemampuan generalisasi yang baik pada dataset real-world. Selain itu, implementasi model dalam bentuk prototipe sistem berbasis web menunjukkan bahwa pendekatan ini tidak hanya efektif secara teoritis, tetapi juga aplikatif dalam mendukung kebutuhan operasional dan pengambilan keputusan pada sistem informasi keamanan.
Kata kunci : Intrusion Detection System, DFNN, false alarm, IoT. SPADEH.

The rapid development of Internet of Things (IoT) technology, particularly in multimedia devices such as IP cameras, has increased the potential for increasingly complex cybersecurity threats. Machine learning-based Intrusion Detection Systems (IDS) have been widely used to identify network attacks, but they still face various challenges, such as suboptimal accuracy, a high false alarm rate, and difficulties in handling high-dimensional data. In addition, the dynamic characteristics of data traffic in IP camera networks require a more adaptive and efficient detection approach. Therefore, a method is needed that can improve the quality of feature representation and the overall performance of the intrusion detection model. This study proposes an intrusion detection approach based on a Deep Feedforward Neural Network (DFNN) optimized using the SPADEH method, which stands for Selective Predictive Attack Detection using Ensemble Hybridization. SPADEH is an ensemble feature selection method that integrates various perspectives on inter-feature relationships through a fusion mechanism to produce a more representative feature subset by considering linear, monotonic, and non-linear relationships among features.The proposed model was evaluated using the public UNSW-NB15 dataset and a real-world dataset obtained from IP camera network traffic. The results show that the proposed approach significantly improves intrusion detection performance compared to the baseline methods. The combination of feature selection methods used proved effective in reducing data dimensionality while improving model accuracy and stability. The optimized DFNN model was also able to reduce the false alarm rate and demonstrated good generalization capability on the real-world dataset. Furthermore, the implementation of the model as a web-based system prototype shows that this approach is not only theoretically effective but also practically applicable in supporting operational needs and decision-making in security information systems.
Keyword : Intrusion Detection System, DFNN, false alarm, IoT, SPADEH

Actions (login required)

View Item