MIT License • Code of Conduct • Contributing Guidelines • Security Policy

ScamShield: Fake Job Posting Detection Platform (2026 Upgrades)

ScamShield is a real-time AI/ML platform that audits job descriptions, checks URLs, verifies company registrations, and analyzes recruiter contact channels to calculate a comprehensive Trust Score (0-100%) before applicants submit their data.

---

🛠️ System Architecture

ScamShield evaluates job postings using a 7-Stage Horizontal Verification Pipeline:

[Ingested] ➔ [NLP Text] ➔ [URL Scan] ➔ [Company Check] ➔ [Recruiter Check] ➔ [Trust Score] ➔ [Decision]

The system verifies credibility across 5 Trust Pillars:

1. NLP Text Analysis (Brain):
   • Evaluates semantics using a machine learning text classifier.
   • Audits vocabulary density and clickbait signals (e.g. excessive exclamation marks).
   • Analyzes lexical complexity via the Flesch Reading Ease index to detect convoluted machine-generated templates.
2. URL Phishing Scan (Link):
   • Inspects embedded links for secure SSL protocols (https:// protocols vs insecure http://).
   • Audits domains against suspicious free hosting providers (e.g., Blogspot, WordPress, Wix) and url shorteners (e.g., bit.ly).
3. Company Check (Magnifying Glass):
   • Validates description completeness (checks if company profiles are missing or too short).
   • Audits company names against official corporate registry structures (identifying standard suffixes like Ltd, Inc, LLP, Pvt).
4. Recruiter Behavior Check (Shield):
   • Scans for personal contact details (embedded personal emails and phone numbers in the description text).
   • Audits email domains, distinguishing between public free email providers (e.g. @gmail.com, @yahoo.com) and validated corporate domains.
   • Cross-references job locations against historical regional fraud ratios.
   • Flags pressure tactics and urgency terms (e.g., "URGENT", "IMMEDIATE").
5. Trust Score Engine (Metric/Dial):
   • Integrates machine learning probabilities and heuristic risk metrics to compute a credibility rating (0 to 100%).
   • 80% - 100%: High Trust (Verified Safe / Approved)
   • 50% - 79%: Neutral Risk (Verification Advised)
   • Below 50%: Low Trust (Scam Vulnerability Alert / Application Blocked)

---

📈 Machine Learning & Model Performance

The core prediction models have been upgraded to maximize prediction accuracy and minimize false-positive warning triggers:

• Text Classifier: Upgraded from CountVectorizer to TfidfVectorizer (with 10,000 max features) and trained an SGDClassifier(loss='log_loss') on balanced training sets using SMOTE. This filters vocabulary noise, raising the individual text F1-score from 76.52% to 80.97%.
• Tabular Classifier: Replaced the linear model with a tree-based RandomForestClassifier (100 estimators) which naturally models non-linear correlations (like location threat ratios, character lengths, and Flesch scores) without scaling bias. This boosted the individual tabular F1-score from 16.54% to 56.32%.
• Ensemble Soft Voting: Combined model outputs using a weighted soft-voting model (0.5 * Text Risk + 0.5 * Tabular Risk) combined with a heuristic check penalty. This produces:
  • Ensemble Accuracy: 98.11% (was 93.74%)
  • Ensemble F1 Score: 84.93% (was 62.31%)
  • Ensemble Precision: 89.86% (was 51.71% — reducing false alarm flags by 90%)
  • Ensemble Recall: 80.52% (was 78.35%)

---

🎨 Premium Visual Interface

The interface is built using Vanilla CSS for clean separation of concerns and features:

• Horizontal Pipeline Tracker: Connects processing nodes with glowing neon tracks, animating step-by-step during live scans.
• Trust Pillars Grid: Sleek, hover-responsive cards representing the 4 checks (NLP, URL, Company, Recruiter) with color-coded status badges (Passed in green, Warning in amber, Failed in red).
• Theme Toggle Switch: Sun/Moon switch in the header that toggles between:
  • Obsidian Dark Theme: A dark obsidian slate background (#080a14) with amethyst purple accents and glowing elements.
  • Alabaster Light Theme: A soft slate-cream background (#f8fafc) with deep indigo highlights and clean card shadows.
• Developer telemetry fine print: A small section at the bottom displaying raw machine learning model flags and location ratios for validation.

---

🚀 How to Set Up and Run

1. Prerequisites

Install the required packages:

pip install flask pandas numpy scikit-learn imbalanced-learn joblib textstat beautifulsoup4 nltk

2. Train the Models

Train and serialize the TF-IDF vectorizers and Random Forest classifiers on the balanced dataset:

python train_model.py

This script will output the accuracy metrics and save the binary files into the models/ directory:

• tfidf_vectorizer.pkl
• clf_log.pkl (text classifier)
• clf_num.pkl (tabular classifier)
• numeric_features.pkl

3. Run the Web Server

Launch the Flask development server:

python app.py

Open your web browser and navigate to http://127.0.0.1:5000 to access the dashboard.

4. Interactive Scrapes

Paste a job listing link (e.g. from LinkedIn or Indeed) into the Smart URL Scan tab. The scraping crawler will auto-fetch the job title, company name, location, and description, pre-populate the input fields, switch to the manual tab, and automatically trigger the ML threat analysis.