JobScan is a machine learning application that detects fraudulent job postings by analyzing both the structure and the language of a listing. Instead of relying on a single signal, the system combines structural cues with the wording of the description itself to flag postings that show patterns typically seen in scams.
The application engineers a set of structural features, including whether the company profile, requirements section, or benefits section is missing, along with the overall description length and the ratio of capital letters used in the text. These are combined with TF-IDF vectorized text using n-grams, which captures the writing patterns and phrasing common in fraudulent listings.
Several classification models were trained and evaluated on the same feature set, including Logistic Regression, Random Forest, Linear SVM, Gradient Boosting, and Multinomial Naive Bayes, to identify the model that best balances precision and recall for this task.
A simple Streamlit interface lets a user paste any job description into a text area and click "Analyze Job Posting" to instantly see whether the listing is likely genuine or fraudulent.
The complete workflow follows an end-to-end pipeline consisting of data cleaning, structural feature engineering, TF-IDF text vectorization, model training and comparison, and real-time prediction through the web interface. This ensures that predictions are grounded in concrete, interpretable signals rather than a single black-box score.
The application can be used by job seekers screening listings before applying, and by job portals or recruitment platforms as a pre-screening layer to flag suspicious postings for manual review.
This project demonstrates the practical implementation of Feature Engineering, Natural Language Processing, and Classical Machine Learning for a real-world trust and safety problem. It was built as a team project, with Harsh leading the overall development effort.