Automated Penetration Testing Using Deep Learning Methods on Wireless Networks

Abstract

Wireless network security is an important aspect of protecting organizational information systems; therefore, penetration testing is necessary to evaluate potential vulnerabilities in network configurations and password policies. This study focuses on conducting penetration testing using Kali Linux with the Aircrack-ng toolset to assess the strength of WPA/WPA2 passwords. One of the main challenges in the password cracking process is the limitation of static wordlists such as common_password.csv, which often fail to capture diverse and contextual password patterns. To address this issue, this research proposes a generative deep learning–based approach utilizing a Long Short-Term Memory (LSTM) architecture. The LSTM model is trained using the common_password.csv dataset to learn character patterns and password structures. The training process includes character tokenization, char-to-index mapping, sequence formation, and training the LSTM network to predict the next character. Once the model is trained, a probabilistic sampling mechanism is applied to generate new password variations that resemble the original dataset distribution while being more diverse. The dynamically generated wordlist is then integrated into penetration testing scenarios using Aircrack-ng to increase the success rate of dictionary-based attacks. The experimental results show that penetration testing using Aircrack-ng with a dataset generated through the LSTM method accurately identified the SSID password, as demonstrated by a testing time ranging from 9 to 12 minutes.