Predictive Modeling of Housing Prices

This project developed a predictive model for housing prices in Charlotte, NC by integrating spatial data on property attributes, neighborhood crime rates, park proximity, and other key features. The model aimed to identify factors influencing home values and uncover spatial patterns that impact property pricing.

To build a robust predictive model that estimates home prices using spatial and non-spatial predictors and to evaluate spatial autocorrelation patterns in residuals to improve model accuracy.

• Home Size and Building Grade: These were the strongest predictors of higher property values, confirming the importance of property characteristics.
• Crime Rates: Areas with higher crime rates showed lower property values, reinforcing the role of safety in home pricing.
• Park Proximity: While closer parks had a slight positive effect, its influence was weaker than anticipated, suggesting other neighborhood features have a stronger influence.
• Year Built: Newer homes generally commanded higher prices, with a clear upward trend for properties built after 1950.
• Spatial Bias: Central neighborhoods were frequently underpredicted, while some suburban areas were overestimated.

Spatial distribution of home prices across Charlotte neighborhoods

Spatial distribution of home sizes (square footage) across Charlotte

Spatial distribution of crime counts across Charlotte neighborhoods

Distance to nearest park across Charlotte neighborhoods

• Higher home prices clustered in suburban regions, where larger properties and newer builds are common.
• Lower property values aligned with areas reporting higher crime counts.
• The presence of residual spatial patterns suggests additional neighborhood-level characteristics may impact pricing.
• Property size (area) showed a strong logarithmic relationship with price, indicating diminishing returns on very large properties.

• Multicollinearity: Variables like home size, bedrooms, and bathrooms had high VIF scores, indicating potential redundancy.
• Outlier Sensitivity: The model struggled to predict high-value homes accurately, likely due to non-linear effects not captured in the model.
• Unaccounted Neighborhood Factors: Unobserved variables such as school quality, walkability, and transit access may contribute to residual spatial autocorrelation.

The model's insights can guide:

• Targeted investment strategies in underpredicted central Charlotte neighborhoods where additional amenities may improve property values.
• Crime prevention initiatives in areas with historically suppressed property values.
• Park improvement efforts in residential areas where green space has stronger correlations with home pricing.
• Housing development policies that consider the strong relationship between property age, size, and market value.

This project demonstrates the power of integrating spatial data and OLS regression in housing price prediction. The combination of engineered features, distance metrics, and spatial diagnostics provided valuable insights into Charlotte's housing market, informing planners and policymakers about factors driving property values. The correlation analysis revealed complex relationships between housing attributes and prices, while the spatial analysis highlighted neighborhood-level patterns that can guide targeted urban development strategies.