Challenge #4 - ML Trained to Infer Investment Areas

[AktivGroupDS]

Summary

I built a classification model to infer if an area is for investment or not. The model trained was XGBoost. I got an Accuracy of 82.29% and 86% for Precision for YES. For NO I got an Accuracy of 82.29% and 78.26% for Precision. These numbers are very good for this type of models.

Approach

My goal was to infer if a location is an Investment Area or not. Using AWS Sagemaker Canvas I worked on some attributes of the table (about 30 attributes). Using the Data Quality functionality I discovered the attributes that have the greatest impact on the model. Then I trained a XGBoost model with the 80% of the data, about 900 rows, and used the 20 for testing. below I am presenting my findings.

Findings

I am surprised to find that the feature (attribute) that has most impact is the median_prop_value and also that any property under $385,000 will be an investment area and any above that value will NOT be.

Some charts from the model

Precision Recall Curve

Confusion Matrix Values

Files

Sagemaker Notebook
SageMakerAutopilotCandidateDefinitionNotebook.ipynb.zip