IntelliPass is an advanced password analysis framework that moves beyond traditional, easily bypassed regular expression (Regex) checks. It implements a Hybrid Thinker architecture, fusing a data-driven Multi-Layer Perceptron (MLP) neural network with a rule-based Mamdani Fuzzy Inference System.
By evaluating credentials from both a statistical pattern-matching perspective and a human-logic expert perspective, IntelliPass provides a nuanced security score that accounts for entropy, character diversity, and known breach patterns.
The system evaluates strength through two distinct "brains":
- The MLP Brain (Data-Driven): A neural network trained on millions of password samples to recognize subtle patterns and common human habits in password creation.
- The Fuzzy Brain (Rule-Based): An expert system using 10+ linguistic rules to judge passwords based on mathematical entropy and character diversity.
- Conservative Scoring: The final verdict is reached using min-pooling—if either model identifies a critical flaw, the final score reflects that weakness to ensure maximum user safety.
- Breach Intelligence: Instantly cross-references inputs against a local database of 14.3 million leaked credentials.
- Similarity Gating: Uses Cosine Similarity to compare new passwords against a history of recently used credentials, preventing predictable "incremental" changes (e.g., changing Summer2025! to Summer2026!).
- Shannon Entropy Analysis: Calculates the cryptographic randomness of the string to ensure it is resilient against dictionary and brute-force attacks.
- Character Heatmaps: Visualizes the "security weight" of each character in the password.
- AI Suggestions: Provides dynamic, context-aware advice on how to improve specific weaknesses identified by the fuzzy rules.
The system transforms raw strings into an 8-dimensional numerical feature vector:
- Length: Total character count.
- Character Ratios: Distribution of Upper, Lower, Digits, and Symbols.
- Shannon Entropy: Measure of informational randomness.
- Pattern Score: Detects sequential characters (abc, 123) and repetitions.
- Diversity Score: Engineered feature representing the richness of the character set domains used.
- Architecture: Two hidden layers (100, 50 neurons).
- Activation: ReLU.
- Optimization: Trained with Adam optimizer and early stopping to prevent overfitting.
- Membership Functions: Trapezoidal functions defining "Short/Medium/Long" and "Low/High Entropy".
- Inference Rules: 10 expert-defined rules that simulate a human security auditor's reasoning.
scikit-learnscikit-fuzzypandas,numpycolorama,lz4,joblib
To re-train the MLP on your local feature dataset:
python train_model.py
To analyze a password and see the full hybrid report:
python predict.py "YourP@ssw0rd123!"
By aggregating the MLP and Fuzzy models, IntelliPass achieves:
- High Interpretability: Unlike "black box" AI, the Fuzzy layer explains why a password is weak.
- High Sensitivity: The inclusion of breach intelligence and entropy analysis ensures a "Very Strong" verdict is only given to truly cryptographic-grade credentials.
Project developed as part of Soft Computing and AI Research.