Home Price Prediction Model

What is my house worth?

In this project, we sought out to predict townhome and single-family home prices in Apex, NC based on square-footage, bedrooms, bathrooms, and a few other things that make a house unique.

Tools: We created a machine learning model utilizing Python, Jupyter Notebooks, and the Tensorflow/Scikit libraries. We also utilized HTML, Javascript/CSS, Flask, and Tableau.
Dataset Used: MLS Triangle Area
Team: Arlette Varela, Nathan Bolt, Alexandra Taft, Anthony English, Nirupama Shankar

Questions:

1. Primary Question: What is my house worth based on selected elements?
2. Secondary Question 1: Is there a better time of the year to sell my house?
3. Secondary Question 2: Can we forecast home prices in Apex for 2021?

Process Summary:

Data Clean-up: Utilized python and tableau to identifity any outliers (such as a home that "sold" for $1.00) and clean the data.

Machine Learning Model Selection:

Considered various models such as random forest, SVC, etc.
Utilized tensorflow sequential because of our number of predictors and need for a numerical outcome.
Chose MeanAbsoluteError to measure our loss because it was a predictive model.
Built two models, one for single family homes and one for townhomes.

Machine Learning Inputs:

We had 9 inputs we used from our data to train our model. Most were numerical already, but we did have to convert Subdivision to a numerical value using label encoder.
Then we had to normalize all of our data inputs to be between 0 -1 to work with the MSE loss function in our model
Below you can see an example of our code that normalized all of our input values before running through the model

Model Accuracy:

R-Square Value/Mean quare Error: Homes/MSE: 0.818/0.006 and Townhomes/MSE: 0.858/0.005
Model Loss: the objective that the model will try to minimize.

Connecting the back-end too the front-end:

Utilized D3 to grab user input and build route url.
Two routes set in app.py: one for single family homes and one for townhomes.
Each ran the input through a different model.
When user selects type of home and selects attributes, route is hit and input is run through respective model, returning predicted value.

Tableau:

Utilized Tableau to answer our secondary questions: 1) Is there a better time of the year to sell my house? and 2) Can we forecast home prices in Apex for 2021?

Website:

Landing Page: Navigation bar consists of Home, Prediction, Tableau Views, and Team sections.

Prediction Section: User selects input, including type of home (single family or townhome) and subdivision.

Prediction Results: Once user clicks "Predict Price", the script will run user input through the model and produce a predicted price.

Tableau Visualizations: You are able to select from 4 graphs for each type of home (Avg price per zipcode, Sales Price/List Price, Days on Market, and Home Size)

Conclusion:

In this project, we were able to successfully build a machine learning model that predicts the values of your home in Apex, NC with an 80-85% accuracy rate. As far as the secondary questions, we utilized tableau to determine that seasonality is a factor you must consider when selling your home as well as to predict future home prices based on previous trends.