A trustworthy security application in the actual world uses the keystroke capability of typing recognition software. Despite being straightforward, it requires a quick and precise method of character analysis. In this manuscript, a keystroke dynamic recognition method to identify and block illegal users is proposed by using deep learning of convolutional neural networks (CNNs) which it can efficiently distinguish legitimate users from illegitimate users. Where, two proposed networks are built based on 1D-CNN to increase and accelerate the recognition abilities. The first network improves the system performance by modifying the kind of activation function utilized, whereas the second network improves the system performance by employing the residual scheme. The findings display that the suggested CNN model with the swish function can deny all illegitimate users with an average equal error rate (EER) of 0.0066. Furthermore, by parallel computing (GPU), the model performance is accelerated by approximately 2 times. Based on the outcomes, the suggested CNN model with swish function significantly outperforms other models in the literature.
Deep Learning Keystroke Dynamic Residual CNN Parallel Computing Authentication System
Birincil Dil | İngilizce |
---|---|
Konular | Endüstri Mühendisliği |
Bölüm | Araştırma Makalesi |
Yazarlar | |
Yayımlanma Tarihi | 29 Temmuz 2024 |
Yayımlandığı Sayı | Yıl 2024 Cilt: 3 Sayı: 1 |