Project 2: Ames Housing Prices Analysis

Problem Statement

• Which features will make for a good investment in Ames

We were tasked with creating a machine learning model using the Ames Housing Dataset. This dataset contained 82 columns of different features and 2,930 observations, 2051 of which were used to train our model.

Executive Summary

I began by reading in and inspecting my data, using various bar chart and heatmaps. During this step, I made sure to analyze the amount of missing values we had in each feature. Next, I analyzed features and their correlations and relationships with price and other features. I then cleaned my data by dropping unneeded columns, dealing with empty values, looking for multicollinearity, eliminating outliers, and finally replacing nominal categorical data with numbers. I added some features and then created my models, incorporating Lasso Regression and Ridge Regression and an analysis of the coefficients they created. Lastly, I explained my insights and conclusions.

Contents:

• Load in and Inspect the Data
• Examine Features
• Clean Data
  • Drop Unneccessary Columns
  • Deal with NaN Values
  • Look for Multicollinearity
  • Remove outliers
  • Replace Nominal Categorical Data with Numbers
• Feature engineering
  • Interaction Features
  • Remove more Features
• Save Clean Data
• Model
  • Read in Clean Data
  • Create X and y
  • Model with Standard Scaler and Ridge with OHE
  • Model with Standard Scaler and Lasso with OHE
  • Categorical Column Transformation with Dummies
  • Model with Standard Scaler and Ridge with Dummies
  • Model with Standard Scaler and Lasso with Dummies
• Conclusion

Data Dictionary

The data dictionary with feature descriptions: http://jse.amstat.org/v19n3/decock/DataDocumentation.txt

Conclusion

Kaggle Model: Both the Lasso and the Ridge model performed very well, but Lasso ended up being the best model with an R^2 score of about .92, suggesting the model explains about 92% of the variance in sales price for the training data.

Business Recommendations: Based on initial analysis and the coefficients of the models used for the Kaggle Competition, it appears homebuyers should look for homes that have a good quality score, have high square footage, and are located good neighborhoods, such as in Northridge, Stone Brook, or Northridge Heights.