Hybrid ML Phishing Detection System
Budget: ₹20,000 – ₹25,000 INR
I’m building PhishVision, a hybrid machine-learning solution that flags phishing sites by combining statistical models with a small, fast rule engine. Your task is to code the core detection module, train and evaluate it, then hand over clean, well-documented Python (or equivalent) code I can drop into a larger security stack.
Key URL features the model must extract and use:
• Length of the URL
• Use of special characters
• The supplied sample URL (https://www.paytm.com/login) should pass through the same parser to prove the pipeline works end-to-end.
Rule-based layer:
• Implement heuristic rules only (no blacklists or regex libraries) to complement the model’s predictions. These heuristics should be configurable so I can tweak thresholds without touching the codebase.
What I expect to receive:
• Feature-extraction script(s)
• A train / test workflow showing accuracy, precision, recall and ROC-AUC on a dataset you provide or point me to
• The lightweight heuristic engine chained after the model with clear comments on which conditions fire and why
• A short README that explains setup, assumptions, and how to extend the rules
I’ll review by running the code against my own unseen URLs and checking that legitimate links stay green while obvious phishing attempts are blocked. Once it passes, the milestone is complete and we can discuss any follow-on work such as browser plug-in integration or live dashboarding.
Key URL features the model must extract and use:
• Length of the URL
• Use of special characters
• The supplied sample URL (https://www.paytm.com/login) should pass through the same parser to prove the pipeline works end-to-end.
Rule-based layer:
• Implement heuristic rules only (no blacklists or regex libraries) to complement the model’s predictions. These heuristics should be configurable so I can tweak thresholds without touching the codebase.
What I expect to receive:
• Feature-extraction script(s)
• A train / test workflow showing accuracy, precision, recall and ROC-AUC on a dataset you provide or point me to
• The lightweight heuristic engine chained after the model with clear comments on which conditions fire and why
• A short README that explains setup, assumptions, and how to extend the rules
I’ll review by running the code against my own unseen URLs and checking that legitimate links stay green while obvious phishing attempts are blocked. Once it passes, the milestone is complete and we can discuss any follow-on work such as browser plug-in integration or live dashboarding.