Cipher-Guard: A Machine Learning Model for Adaptive and Context-Aware Password Security

Abstract

This research aims to enhance password security by designing, training and testing Cipher-Guard, a machine learning (ML) algorithm that incorporates complex features and techniques derived from proven feature engineering. The current model is inherently built on the equations of mathematical modelling concerning complexity metrics of passwords and contextualisation. This is in terms of password length (PL), entropy (E) and the character set diversity (CSD) of the passwords, as well as contextual embeddings such as user-specific password history (UPH) and temporal patterns (TPs). A simulation of the data matrix was created, containing 1000 records, which formed the basis of the model and a thorough exploratory data analysis (EDA). The proposed model demonstrates interpretability and credibility in distinguishing patterns related to password complexity metrics, yielding promising formative results in the specified evaluation metrics. PL, E and CSD were analysed thoroughly, and the importance of increasing password security was identified. Specific contextual embeddings were identified as UPH and TPs, which reflect the model’s ability to adapt to the particular user’s actions. Including adversarial training features also helped protect the model from potential manipulations through a proactive defence plan. Moreover, the enhancement of cryptographic principles, which include hashing and salting, also improved the dataset security, thereby increasing the standard practice within the industry. The metrics involved in the comparative assessment included receiver operating characteristic-area under the curve (ROC-AUC), learning curves, confusion matrix, precision matrix and precision-recall curves, among others. Cipher-Guard performed consistently well across these parameters, which has reinforced its authenticity in strengthening password protection. However, to appreciate such findings, certain constraints need to be well addressed, such as the quality of the chosen dataset, some ethical concerns raised in the study and the fact that new threats are constantly emerging. Recommendations for enhancing continuous dataset monitoring and establishing an ethical framework for the case are also suggested. The research directions broaden future possibilities for research while emphasising continuous model updating, individual-centred protection and compatibility with other emerging technologies. Therefore, Cipher-Guard represents a significant advancement in password protection and offers a promising prospect for flexible and personalised security in the modern era.