A Password Strength Classifier Using Supervised Machine Learning

Having a strong password is one of the basic protections one can have against identity theft and hacking. Yet, data gathered about annual security breaches suggests that accounts being compromised due to weak passwords amounts to about 61% of all security breaches. Polls and surveys conducted in first world countries show us that a vast majority of users do not frequently reset their passwords, reuse passwords across multiple sites and keep weak passwords to begin with. This scenario will be even worse in countries like ours where technical knowledge among the general public is limited. Through this work, we aim to make a password classifier which would tell the users the strength of their entered passwords, thereby, highlighting any potential weaknesses in their passwords. Here, we have used multiple classifier models to test the strength of passwords including Logistic Regression, Naive Bayes, KNN, Random Forest and XGBoost. Their performance metrics are tested and a maximum accuracy of 0.9933 is achieved along with a Matthew's Correlation Coefficient of 0.99. XGBoost is selected a the best-performing classifier for this case. With highest value of accuracy (0.9933), Matthew's Correlation Coefficient (0.99) and AUC (0.9953). To cater the need for strong passwords, we also proposed a password generator that generates strong password.