🧠 Income Classification with Census Data

This project implements predictive models to classify adult income based on the U.S. Census Bureau data. The goal is to predict whether a person earns more than $50K per year using machine learning techniques.

---

📁 Files in This Repository

File	Description
FinalProject.ipynb	Jupyter Notebook with complete code and outputs
README.md	Project overview and usage instructions
adult-dataset.csv	The data is in the file "adult-dataset.csv". It was extracted from the census bureau database, found at: http://www.census.gov/ftp/pub/DES/www/welcome.html

---

📊 Dataset Overview

The dataset contains demographic and employment-related attributes for U.S. adults. The target variable is Income:

• <=50K or >50K

Features include:

• Age (int)
• Work Class (categorical)
• Education (categorical)
• Marital Status (categorical)
• Occupation (categorical)
• Race (categorical)
• Sex (binary)
• Hours per week (int)

---

🧹 Part A: Data Cleaning

• Handled missing data by removing or imputing invalid entries
• Converted categorical variables using one-hot encoding
• Removed irrelevant columns
• Ensured numeric data for model compatibility

---

📉 Part B: Dimensionality Reduction

Applied both SVD and PCA to reduce dataset dimensions.

✅ Results:

• Explained Variance: PCA components capturing >90% of varianc

🤖 Part C: Model Training & Evaluation

1. Multi-Layer Perceptron (Neural Network)

• Built with MLPClassifier
• Evaluated using Confusion Matrix & Classification Report

2. Logistic Regression

• Applied using Scikit-learn
• Good baseline model

3. Naïve Bayes (GaussianNB)

• Fast and interpretable model

4. K-Means Clustering

• Unsupervised clustering excluding Income

---

🧪 Results Summary

Model	Accuracy	Precision	Recall	F1 Score
MLP	✅ High	✅ High	✅ High	✅ High
Logistic Regression	Moderate	Moderate	Moderate	Moderate
Naïve Bayes	Moderate	Lower	Moderate	Moderate
K-Means	-	-	-	- (unsupervised)

---

🚀 Getting Started

🔧 Install Dependencies

pip install numpy pandas matplotlib seaborn scikit-learn