Predictive Model Analysis (2011 Predictors on 2013 Market Value). Using Thads Dataset

 

Stakeholder Report: Predictive Model Analysis (2011 Predictors on 2013 Market Value)

Problem Statement

The primary business objective is to understand and predict the market value of properties. Specifically, this analysis aims to determine how well observable housing characteristics from 2011 can predict the natural logarithm of the Market Value (LN(VALUE)) in 2013. Accurate predictions can assist in strategic planning, investment decisions, and market valuation assessments.

1. Data Importation

Objective: To establish a foundational dataset for predicting 2013 property values.

  • Method: Two datasets, Thads 2011.txt and Thads 2013.txt, containing housing characteristics and market values respectively, were imported from Google Drive into the Colab environment using pd.read_csv().

2. Data Wrangling and Preprocessing

This phase was critical to transform raw data into a clean, normalized, and model-ready format.

  • Initial Merging & Filtering: Relevant variables from 2011 (e.g., FMR, BUILT, UTILITY, BEDRMS) and 2013 (VALUE) were selected and merged based on the CONTROL identifier. Data was then filtered to include properties with FMR >= 1, Value_2013 >= 1000, and Utility > 0 to ensure data quality.
  • Logarithmic Transformation: The depende…ogarithms (LN_Value_2013, LN_FMR, LN_Utility). This helps to normalize skewed distributions and allows for the interpretation of coefficients as percentage changes.
  • Categorical Variable Encoding: Categorical variables (Metro, Status, Type) were first re-categorized (e.g., 'Urban'/'Rural', 'Occupied'/'Vacant') for clarity and then converted into numerical format using one-hot encoding (pd.get_dummies()). The resulting boolean columns were converted to integers (0s and 1s).

3. Analysis (Exploratory Data Analysis)

Exploratory analysis provided insights into data distributions and relationships.

  • Correlation Heatmap: A correlation matrix and heatmap were generated to visualize relationships between all processed variables. This helped identify potential multicollinearity and the strength of relationships with LN_Value_2013.
    • LN_Value_2013 showed the strongest positive correlations with LN_FMR (0.518), FMR (0.509), and Bedrooms (0.320). I… Utility (0.269) and LN_Utility (0.239).
    • High correlations were observed between FMR and LN_FMR (0.978), and Utility and LN_Utility (0.905), indicating potential multicollinearity among these original and log-transformed pairs.


  • Line Chart: Built vs FMR: This visualization showed the trend of Fair Market Rent (FMR) over the years properties were built, including its log-transformed version. This helped understand the temporal relationship between property age and FMR


4. Machine Learning Model

…Linear Regression Model Metrics (on Test Set):

  • Mean Absolute Error (MAE): 0.5464
    • Interpretation: On average, the model's prediction for the natural logarithm of 2013 property values deviates from the actual LN_Value_2013 by 0.5464 units. To interpret this in dollar terms, one would need to exponentiate this value. This indicates the typical magnitude of prediction error in the log-transformed scale.
  • R-Squared: 0.2976 (approx. 0.30)
    • Interpretation: Approximately 30% of the variance in the natural logarithm of 2013 property values can be explained by the 2011 housing characteristics included in the model. This is a more realistic R-squared value after addressing the data leakage and suggests that the 2011 features have a moderate predictive power for 2013 values.

OLS Model Summary (for detailed insights into coefficients):




Interpretation of OLS Coefficients & Business Impact:
  • FMR (Coefficient: 0.0004, p-value: 0.000): Highly significant. A small positive change in 2011 Fair Market Rent is associated with an increase in 2013 property value. This implies that properties in areas with higher rental demand in 2011 tended to appreciate or hold their value better by 2013. Business Impact: FMR can be a useful leading indicator for property investment strategies.
  • **Built (Coefficient: 0.0033, p-va…egression models such as Ridge or Lasso Regression (for regularization and handling multicollinearity), or potentially non-linear models if relationships are found to be non-linear.
  1. Outlier Detection and Treatment: Analyze the data for outliers that might be disproportionately influencing the model and consider appropriate treatment methods.
  2. Model Validation: Perform more rigorous cross-validation techniques to ensure the model's robustness and generalization to new, unseen data


Comments

Popular posts from this blog

Teach your Child Coding

Google Data Analytics Project 2: Bike Share Case Study Project

ANALYZING CHANGE IN MEAN MARKET VALUE THADS DATASET