predict-housing-prices

The project is done in R.

The goal of the project is to build a model to predict housing prices in two markets, Pittsburgh, PA and Richmond, VA, using a scraped data set.
The delivarables include:

Predictions: a single .csv file with 600 housing prices predictions.
Non-Technical Report: a 1 page long .pdf file that discusses the findings and implications tailored for a non-technical desicion maker.
Technical Report: a 4 pages long .pdf report that outlines the process, the decisions behind certain cleaning and preprocessing choices, model selection, etc. for a technical audience.
Code: an .Rmd file with the code including exploratory data analysis, preprocessing, and models' trainings.

The metric used for acessing the performance was MSE. Random Forest performed best.

0. Load Data

The dataset is 1400x16. There are both continious and categorical variables, including missing data.

1. Exploratory Data Analysis & Visualizations

Checking the dimentions, getting a glimpse of the data, checking the datatypes, summaries, variables with near-zero variance, unique values, etc.

2. Preprocessing

Please see the Reports for the details of preprocessing decisisons.

3. Training Models

Linear and Non-Linear Models
- LM with various variables (3, 7) based on the following selections:
  Best Subset Selection, Forward Stepwise Selection, Backward Stepwise Selection
- Ridge Regression
- Lasso Regression
- Principle Component Regression (PCR)
- Partial Least Squares Regression (PLS)
- Polynomials
- Splines
Tree-based Models
- Trees
- Bagged Tree
- Random Forest
- Boosted Trees
K-Nearest Neighboor (KNN)

4. Choose the best model

Mean Squred Error (MSE) was used to compare perfomance of the models.
Lowest MSE was used to choose the best model. RF performed the best.