Boston Housing Price Prediction - Linear Regression
MY ROLES
TIMELINE
2019
TOOLS USED
Linear Regression, Statsmodels, Scikit-learn, Python, Pandas, Seaborn, Outlier Detection, Residual Analysis, Feature Engineering, Statistical Modeling
DESCRIPTION
This project implemented linear regression models to predict housing prices using the classic Boston housing dataset. Beyond basic modeling, the analysis focused on improving model performance through systematic identification and removal of outliers and high leverage points. The comparative analysis demonstrated significant improvements in model quality as measured by F-statistics and AIC scores.
OUTCOMES
- Developed multiple linear regression models with progressive improvements in predictive performance
- Demonstrated significant model enhancement through outlier and high leverage point removal (F-statistic: 244.2 → 366.2)
- Achieved substantial reduction in AIC scores (3233 → 3010) indicating better model quality
FEATURES
Comparative model evaluation, outlier detection and removal, leverage point analysis, residual diagnostics, model quality metrics analysis
CHALLENGES
Identifying meaningful outliers versus legitimate data points. Balancing model complexity with interpretability. Determining which observations had undue influence on regression coefficients.
APPROACH
Implemented multiple linear regression models using statsmodels and scikit-learn. Conducted systematic analysis comparing four model variations: with outliers, without outliers, without high leverage points, and without both outliers and high leverage points. Evaluated each model using F-statistics and AIC, demonstrating that the model without outliers and high leverage points achieved the best performance (F-statistic: 366.2, AIC: 3010).