Predicting Password Strength
The project includes code of a model that predicts password strength. The implementation is rather simple and naive. It is inspired from this paper
- length of the password
- is the password a common password
- does it contain numbers
- does it contain special characters
- does it contain block letters
The model achieves an accuracy of 99.9% on test data.
The model is not robust to the real world standards. For an instance, the model assigns “password” a strength of 1. The dataset is small and doesn’t contain enough common passwords, neither is it personalized to the user.
- Meta learning can be useful in the domain to help make a model personalizable to the user.
- tfidf scores can be used to evaluate the model
- A better dataset is always a possibility
- More hand-picked parameters like position of occurence of special characters
I’ve included a common passwords file (taken probably from wikipedia)
dehasher.py, we can test hashcat ability to obliterate common passwords.