Adult Census Income Prediction

A binary classification project that predicts whether a person earns >50K or ≤50K per year based on demographic and employment data from the 1994 US Census.

Three ML models are compared through a single end-to-end sklearn pipeline — Logistic Regression, Gradient Boosting, and Random Forest — with Random Forest coming out on top at 85.65% test accuracy and ROC-AUC of 0.9097.

---

Dataset

Source: UCI Adult Census Income Dataset (Kohavi & Becker, 1994 US Census Bureau)

Split	Rows
Training	32,561
Test	16,281

Features (14): age, workclass, education, education-num, marital-status, occupation, relationship, race, sex, capital-gain, capital-loss, hours-per-week, native-country

Target: income → >50K (1) or ≤50K (0)

---

Project Workflow

Step 1 → Import Libraries
Step 2 → Load Dataset
Step 3 → Exploratory Data Analysis (EDA)
Step 4 → Data Preprocessing
Step 5 → Build ML Pipeline
Step 6 → Train the Model
Step 7 → Evaluate (Metrics + Confusion Matrix + ROC Curve + Feature Importance)
Step 8 → Save & Reload Model (joblib)
Step 9 → Predict on New / Batch Data

---

Tech Stack

• Language: Python 3
• Environment: Google Colab
• Libraries: scikit-learn, pandas, numpy, matplotlib, seaborn, joblib

---

Pipeline Architecture

The entire preprocessing and modeling logic is wrapped in a single sklearn.pipeline.Pipeline.

ColumnTransformer
├── Numerical: SimpleImputer (median) → StandardScaler
└── Categorical: SimpleImputer (most_frequent) → OneHotEncoder

Pipeline
├── preprocessor: ColumnTransformer (above)
└── classifier: RandomForestClassifier / GradientBoostingClassifier / LogisticRegression

---

Results

Model	Test Accuracy	ROC-AUC
Logistic Regression (baseline)	84.10%	0.88
Gradient Boosting	85.02%	0.90
Random Forest	85.65%	0.9097

Random Forest config: n_estimators=200, max_depth=10, min_samples_leaf=5, random_state=42

Cross-validation: 5-fold stratified CV on training set

Top predictors (by feature importance):

1. marital-status (married-civ-spouse)
2. capital-gain
3. education level

---

Files

├── Predict_Adult_Income.ipynb           # Main notebook (full pipeline)
├── adult.data                           # Training set
├── adult.test                           # Test set
├── adult_income_pipeline.pkl            # Saved model (joblib)
├── Adult_Income_ML_Project_Report.pdf   # Full project report
└── Comparative_Performance_Analysis...pdf  # Research paper format

---

How to Run

1. Open Predict_Adult_Income.ipynb in Google Colab
2. Upload adult.data and adult.test when prompted
3. Run all cells top to bottom
4. The trained pipeline gets saved as adult_income_pipeline.pkl and downloaded automatically

---

Sample Prediction

import joblib, pandas as pd

pipeline = joblib.load("adult_income_pipeline.pkl")

new_person = pd.DataFrame([{
    "age": 35, "workclass": "Private", "education": "Bachelors",
    "education_num": 13, "marital_status": "Married-civ-spouse",
    "occupation": "Exec-managerial", "relationship": "Husband",
    "race": "White", "sex": "Male", "capital_gain": 5000,
    "capital_loss": 0, "hours_per_week": 45, "native_country": "United-States"
}])

prediction = pipeline.predict(new_person)
probability = pipeline.predict_proba(new_person)
print(">50K" if prediction[0] == 1 else "<=50K")

---

Key Takeaways

• Ensemble methods (Random Forest, Gradient Boosting) outperform linear classifiers on this structured tabular dataset
• A clean sklearn pipeline makes preprocessing fully reproducible and deployment-ready
• Marital status, capital gain, and education are the strongest income predictors in this dataset

---

References

• Kohavi, R. (1996). Scaling Up the Accuracy of Naive-Bayes Classifiers: A Decision-Tree Hybrid. KDD.
• UCI ML Repository: Adult Dataset
• Breiman, L. (2001). Random Forests. Machine Learning, 45(1), 5–32.
• Friedman, J.H. (2001). Greedy function approximation: A gradient boosting machine. Annals of Statistics.