Detail Cantuman

Perkembangan cloud computing telah membawa kemudahan dalam pengelolaan data dan layanan berbasis jaringan, namun juga menghadirkan tantangan keamanan yang kompleks, terutama terhadap ancaman anomali lalu lintas akibat serangan Low-Rate Distributed Denial of Service (Low-Rate DDoS). Serangan ini sulit dideteksi karena memiliki pola lalu lintas yang menyerupai akses normal tetapi tetap dapat menurunkan kualitas layanan. Penelitian ini mengusulkan pendekatan deteksi serangan Low-Rate DDoS dengan mengkombinasikan seleksi fitur Analysis of Variance (ANOVA) dan algoritma Naive Bayes untuk mengukur akurasi klasifikasi serta efisiensi dalam proses analisis data. Dataset CICIDS2017 digunakan untuk pelatihan dan pengujian model dengan proses seleksi fitur bertujuan mengukur akurasi serta efisiensi klasifikasi. Hasil penelitian menunjukkan bahwa meskipun akurasi pada K=10 mencapai 74%, pemilihan fitur optimal pada K=10 memberikan keseimbangan terbaik antara efisiensi dan akurasi deteksi. Pemisahan kelas Benign dan Attack pada confusion matrix pada K=10 menunjukkan distribusi yang lebih seimbang, menjadikannya pilihan terbaik dalam mendeteksi serangan Low-Rate DDoS secara efektif.

The development of cloud computing has brought convenience in managing data and network-based services, but it also presents complex security challenges, especially against traffic anomaly threats and Low-Rate Distributed Denial of Service (Low-Rate DDoS) attacks. These attacks are difficult to detect because they have traffic patterns that resemble normal access but can still reduce service quality. This study proposes an approach to detect Low-Rate DDoS attacks by combining Analysis of Variance (ANOVA) feature selection and the Naive Bayes algorithm to measure classification accuracy and efficiency in the data analysis process. The CICIDS2017 dataset was used for model training and testing with a feature selection process aimed at measuring classification accuracy and efficiency. The results show that although the accuracy at K=10 reaches 74%, the optimal feature selection at K=10 provides the best balance between efficiency and detection accuracy. The separation of the Benign and Attack classes in the confusion matrix at K=10 shows a more balanced distribution, making it the best choice for effectively detecting Low-Rate DDoS attacks.