Stroke Predictions Dataset

Summary

This dataset was imported, cleaned, and visualized. The purpose of this is to help create a model that can determine if a patient is likely to get a stroke based on the metabolic parameters provided.

Data Dictionary

Column Name	Data Type	Description
id	Integer	Unique identifier
gender	Object	"Male", "Female", "Other"
age	Float	Age of patient
hypertension	Integer	0 if the patient doesn't have hypertension, 1 if the patient has hypertension
heart Disease	Integer	0 if the patient doesn't have any heart diseases, 1 if the patient has a heart disease
ever_married	Object	"No" or "Yes"
work_type	Object	"children", "Govt_jov", "Never_worked", "Private" or "Self-employed"
Residence_type	Object	"Rural" or "Urban"
avg_glucose_level	Float	average glucose level in blood
bmi	Float	body mass index
smoking_status	Object	"formerly smoked", "never smoked", "smokes" or "Unknown"*
stroke	Integer	1 if the patient had a stroke or 0 if not, target

*Note: "Unknown" in smoking_status means that the information is unavailable for this patient

Data Cleaning

• Fixed inconsistency of values

Data Visualization

• Created plots to determine what factors affected chances of stroke

Machine Learning Model

• Median Imputer used for numerical values
• OneHotEncoded used for categorical values
• Data was scaled

Model Comparison

• KNN Classifier
• Random Forest Classifier
• XGBooster

Accuracy Comparisons

	KNN Classifier	Random Forest Classifier	XGBooster
Baseline - Training	96%	100%	96%
Baseline - Testing	94%	94%	96%
Hypertuned - Training	96%	96%	96%
Hypertuned - Testing	94%	94%	94%