SalaryPredict - README

📌 Overview

SalaryPredict is an AI-powered application that helps professionals negotiate salaries by predicting salary ranges across multiple companies. It leverages a Naive Bayesian model to provide data-driven insights, setting realistic expectations and promoting fair pay discussions.

🚀 Features

• Accurate Predictions: Uses a Naive Bayesian model trained on 5,000+ records with 85% accuracy.
• Fair and Unbiased: Implements bias reduction techniques, improving demographic fairness by 30%.
• Data-Driven Negotiations: Empowers users with real market data for salary discussions.

Implementation Details

Constructing the Bayesian Network:

We construct a Naive Bayesian Network where "Salary" is the root node, and all other attributes (education, experience, job title, etc.) are conditionally dependent on it. The network is built using data from adult-train.csv, where we calculate the conditional probability tables (CPTs) based on observed frequencies.

Variable Elimination:

Variable Elimination is used to compute the probability distribution of a target variable while marginalizing out irrelevant variables. The process follows these steps:

1. Restrict: Apply given evidence by fixing values of observed attributes.
2. Multiply: Combine factors to represent joint distributions.
3. Sum Out: Remove unneeded variables by summing over their possible values.
4. Normalize: Ensure that probabilities sum to 1.

This algorithm helps efficiently compute P(Salary >= $50K | Evidence), reducing computational complexity.

Salary Prediction:

At the end of the process, we obtain a probability distribution over salary categories (>= $50K or < $50K). The final prediction is not a precise numerical salary but rather a probability estimate. If P(Salary >= $50K | Evidence) > 0.5, the model predicts a high salary; otherwise, it predicts a low salary. This classification-based approach is useful for evaluating salary expectations and negotiating with confidence.

Fairness Evaluation:

• Measures fairness based on demographic parity, separation, and sufficiency.
• Uses evidence sets (E1 and E2) to analyze fairness metrics.
• Reduces demographic bias by 30% for more equitable salary predictions.

Data Files

• adult-train.csv: Training data.
• adult-test.csv: Test data.
• naive_bayes_solution.py: Implements the Bayesian model.
• bns.txt: Contains fairness evaluation answers.
• autograder.py: Basic tests to verify correctness.

How to Run

1. Ensure Python 3 is installed.
2. Place all necessary files in the same directory.
3. Run python naive_bayes_solution.py to train and test the model.
4. Run python autograder.py to verify the implementation.
5. Review the output for fairness analysis and salary predictions.