Analisis Ketahanan Algoritma Enkripsi KTANTAN64 terhadap Serangan Berbasis Deep Learning

Abstract

Enkripsi merupakan proses transformasi pesan asli (plaintext) menjadi teks sandi (ciphertext) untuk menjaga kerahasiaan informasi sehingga ketahanan algoritmanya perlu dievaluasi. Penelitian ini bertujuan mengembangkan model deep learning untuk mensimulasikan serangan pemulihan plaintext tanpa kunci pada algoritma KTANTAN64 dan menganalisis ketahanan algoritma tersebut berdasarkan jumlah ronde enkripsinya. Metode deep learning yang diusulkan menggunakan arsitektur Bidirectional Long Short-Term Memory (BiLSTM) dengan Gated Linear Unit (GLU) serta teknik bipolarisasi data. Hasilnya, model mencapai Bit Accuracy Probability (BAP) 91,09% dan Character Accuracy Probability (CAP) 66,27% pada 50 ronde enkripsi yang memiliki tingkat pengacakan 24,26%. Pada ronde maksimal (254 ronde), pengacakan menyentuh kriteria ideal 50,01% sehingga BAP turun ke angka 89,91% dan CAP tertahan di 65,05%. Hal ini membuktikan peningkatan efek pengacakan menurunkan performa model. Dengan demikian, algoritma KTANTAN64 tangguh terhadap serangan pemulihan plaintext, terutama pada jumlah ronde maksimal.

Encryption is the process of transforming an original message (plaintext) into ciphertext to maintain information confidentiality, making it necessary to evaluate its algorithmic resistance. This study aims to develop a deep learning model to simulate plaintext recovery attacks on the KTANTAN64 algorithm and analyze the algorithm's resistance based on the number of its encryption rounds. The proposed deep learning method employs a Bidirectional Long Short-Term Memory (BiLSTM) architecture with a Gated Linear Unit (GLU) and a data bipolarization technique. The model achieves a Bit Accuracy Probability (BAP) of 91.09% and a Character Accuracy Probability (CAP) of 66.27% at 50 encryption rounds, which exhibit an avalanche effect of 24.26%. The maximum rounds (254 rounds), the avalanche reaches the ideal criterion of 50.01%, causing the BAP to drop to 89.91% and the CAP to plateau at 65.05%. This proves the increase in the avalanche effect degrades the performance of the model. Thus, the KTANTAN64 algorithm is robust against plaintext recovery attacks, especially at the maximum number of rounds.