Adult Data Set

Dataset Overview

Data Source and Purpose

• Source: UCI Machine Learning Repository
• Name: Adult Data Set
• Purpose: Predict whether an individual's annual income exceeds $50K/year based on provided census data.

Data Processing Steps

1. Data Import: Use pandas read_csv function to read the "adult.data.txt" file.
2. Feature Labels: Assign labels to the file’s features, including Age, Workclass, fnlwgt, Education, Education_Num, Martial_Status, Occupation, Relationship, Race, Sex, Capital_Gain, Capital_Loss, Hours_per_week, Country, Target.
3. String Data Conversion: Convert string variables such as gender to numeric values (e.g., Female → 0, Male → 1).
4. Missing Value Handling: Use fillna function to fill NaN values.
5. Feature and Target Selection: Select features and target via data[features].values and data["target"].
6. Dataset Size Display: Show the number of records with X.shape[0].

Model Training and Evaluation

• Initial Model: Train a logistic regression model, adjusting hyperparameters such as the C value.
• Optimization Algorithm: Apply GridSearchCV for hyperparameter tuning, testing different penalty (l1, l2) and C values (0.01, 0.1, 1, 10, 100, 1000) to find the optimal model.
• Model Comparison: Compare traditional logistic regression, GridSearchCV‑optimized logistic regression, and k‑nearest neighbors algorithm.
• Evaluation Results: Present a table of precision, recall, f1‑score, and support for each model, e.g., GridSearchCV‑optimized logistic regression results:
  • Under 50k: precision=0.87, recall=0.93, f1-score=0.90, support=7407
  • Over 50k: precision=0.72, recall=0.58, f1-score=0.64, support=2362
  • Average/Total: precision=0.84, recall=0.85, f1-score=0.84, support=9769

Conclusion

After a series of preprocessing steps, the dataset is used to train and evaluate logistic regression models, and GridSearchCV optimization of parameters significantly improves model performance.