Data Patterns in Ames Housing
The skills I demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
Ames Housing
There are many many parties who are greatly interested in accurate understanding of house prices for reasons both personal and financial. Therefore it is a subject of immense value to create good (or in some cases better) models for house prices, as well as deepening our conceptual understanding of the housing market in general.
We use various ML techniques to model the Ames, IA, housing price data, thereby considering different issues and tackling different problems associated with the same data set.
Neighborhood-based Demographic Visualization
We do the usual data cleaning and imputation that you would typically do for any data analysis project.
But in addition, we also do an exploratory analysis of Ames housing development. Specifically, we do a neighborhood and distance-based analysis to aggregate Ames houses and create a visualization of construction trends across over time and how they overlay neighborhoods and price.
House Values in different geographical areas
Data Patterns in Ames Housing
Linear Models
We use a generic multi-linear correlation model (implicitly without regularization) and compare that to a LASSO model with L1 penalties. Specifically, we compare the performance of each model on train/test sets to evaluate the possibility of overfitting.
The important features from the lasso regression is below
Data Patterns in Ames Housing
Tree Models
We also consider a generic Random Forest model and Gradient Boosting model along with a modified Random Forest model with a Term Structure adjustment. We evaluate the Random Forest vs. the Gradient Boosting model in terms of accuracy and feature importance, and we evaluate the Term Structure adjustment in terms of ensemble improvement to the Random Forest.
About Authors
john kosmicke
View all posts by john kosmicke >