This project focuses on accurate predictions of house prices in Ames, Iowa. The dataset used is a subset of the Ames Iowa Housing Dataset from the Ames City Assessor's Office, made available by Dean De Cock of Truman University in 2011. Professor De Cock's original paper discussing the data can be found here.
Presentation Slides for this project can be found here.
This project aims to create business value for real estate agents, buyers, sellers, and home-flippers in Ames, Iowa by:
- predicting property sale values
- to help sellers appropriately valuate and list their homes
- to help buyers make informed bidding decisions, and recognize underpriced or overpriced homes
- providing insights into what kinds of renovation recommendations should be made to clients looking to sell their home, or clients looking to buy a home to flip
- providing insights into what kinds of static house features potential buyers should prioritize, to maximize the potential resale value of their home - eg. Neighborhood, Lot Shape etc
- Regression
- EDA/Univariate Regression
- Multiple Linear Regression
- Lasso Regression
- Elastic-Net
- Tree-based Models
- Decision Tree
- Random Forest
- Boosting
- Support Vector Regression
For linear regression, features were selected if they were:
- significant (p-values)
- not multicolinear (did not inflate VIF of other features)
For Tree-based models, features were selected using Lasso Regression.
- The code for this project is split across the following notebooks:
I. EDA & Feature Engineeringcontains data cleaning, exploratory analysis, and first round feature engineeringII. Preprocessing, Modeling and Evaluationcontains second round feature engineering/selection, preprocessing, model fitting and evaluation code for linear and tree-based models.helper_modulecontains functions to help with EDA and model evaluation
- Data (raw and generated) is in
data - Figures for presentation are in
figures