Data Analysis on Valuable House Features
The skills I demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
Introduction
If you're an investor looking to flip houses in Ames IOWA we will evaluate the data on the house features that will bring the most return on your investment. Our objective is to come up with real numbers for each recommended feature.
Our estimations are based on the housing data for the houses in Ames, IOWA sold for 2006-2010 (obtained from Kaggle). We cleaned and prepared the data to build a good model with it. A good model is essential in our case as we are going to use it extensively in our prediction of a gain that a potential investor will incur from each feature.
Data cleaning
The data consists of 2,580 observations and 81 features. After vetting the data for duplicates we were left with a total of 2,579 houses. We examined the columns and had to dismiss a few that do not bring value to our study ('MiscFeature', 'Alley') or could potentially skew the data ('PoolQC'). We inspected the dataset for multicollinearity - did not exclude any features in the result. We also carefully examined 'SaleCondition' feature and eliminated any 'abnormal' sales such as foreclosure etc. These sales account for 6% of total observations.
Data on Feature Selection and Model Creation
We ran the prepared dataset through the multiple linear model. The data shows overfitting as the train and test data resulted in .95 and .93 R-squared respectively. To determine feature importance we used Lasso. It returned 15 features.
Lasso selected features and corresponding coefficients:
Next, we evaluated p-values and R-squared using Statmodels. With the features selected by Lasso the model returned .92 and .91 R-squared for its train and test set. We left only 7 features for our final model, which returned .92 for both train and test set:
Data Feature Evaluation
With 'Overall Quality' and 'Overall Condition' being in our main set of features that need renovation, we determined several additional features for evaluation: Kitchen Quality, Garage Finish, Exterior Finish, Basement, Heating and Central Air . We used coefficients to determine the return on investment. For the evaluation technique/method please refer to our GitHub
Kitchen Quality
The initial EDA has shown that Kitchen renovation would contribute to the sale of the house the most. The graph below shows how Kitchen Improvement contributes to the value of the house both percentage wise and in dollar amount. The 'Average' kitchen yields a 1.6% on the house. The 'Good' kitchen brings in 3.8% and 'Excellent' 8.7%. Also, please check the dollar amount on the y-axis.
Garage Finish
Garage Finish is the interior finish of a garage. Garage upgrade does not signify much difference in price. However there's still significant difference in value of a house with 'Finished' garage:
Exterior Finish
We evaluated different finish material. The graph below shows our findings. If you redo your Plywood home and give it the Vinyl finish, you'll have 4.2% return on your investment. However, we would recommend it only if the plywood on your home is in bad condition. It's worthy mentioning that Vinyl brings better return than Metal Siding or Wood Siding.
Basement
In case of Basement, prices drop when it comes to "Excellent" basement condition instead of going up. However, they do go up a little when we compare "Average" Basement with "Good" basement.
Heating & Central Air
The initial EDA shows that most homes have heating at 'Good' and higher.
Therefore, we skipped our calculations on Heating as it's cold in IOWA as it goes without saying that Heating is a certain return on investment.
Same with Central Air. 94% of the house have a good Central Air.
About Authors
Khamanna Iskandarova
View all posts by Khamanna Iskandarova >
Ariel Perez
View all posts by Ariel Perez >
Shalabh Agarwala
View all posts by Shalabh Agarwala >