Stroke-Prediction-ML-Model

프로젝트 주제

환자들의 뇌졸중을 예측하는 이진 분류 모델 학습

프로젝트 목적

뇌졸중 고위험 환자들을 구별하여 추가적인 검사를 통해 조기에 정밀치료 및 예방

데이터 설명

train_strokes.csv - 43000 환자의 임상 정보

• 뇌졸중을 경험하지 않은 환자 (98.2%), 뇌졸중을 경험한 환자는 783명 (1.8%) 으로 불균형한 데이터

id, gender, age, hypertension, heart_disease, ever_married, work_type, Residence_type, avg_glucode_level, bmi, smoking_status, stroke

프로젝트 정리

뇌졸중 경험 유무를 target variable로 설정 후 다양한 모델들을 사용하여 GridSearchCV를 통해 학습 및 결과 확인

• 불균형한 데이터 분포 문제를 해결하기 위해 SMOTE과 SMOTEENN resampling 방법들을 사용

models: LogisticRegression, DecisionTreeClassifier, RandomForestClassifier, XGBClassifier, AdaboostClassifier, GaussianNB

< 이미지 출처: A Comparison of Undersampling, Oversampling, and SMOTE Methods for Dealing with Imbalanced Classification in Educational Data Mining >

프로젝트 결과

• AUC 점수, 재현율(Recall), F1 점수, 정확도(Precision)를 지표로 사용하여 모델의 성능을 비교
• 뇌졸중 고위험군인 환자를 예측하는 것이 주목적이므로 재현율(Recall) 점수에 집중
• resampling을 진행한 모델의 결과 향상

Model	AUC Score	Recall	F1 Score	Precision
LogisticRegression (SMOTEENN)	0.8374	0.7565	0.4740	0.5233
LogisticRegression (SMOTE)	0.8356	0.7554	0.4729	0.5230
GaussianNB (SMOTE)	0.8031	0.7384	0.4543	0.5199
GaussianNB (SMOTEENN)	0.8064	0.7365	0.4632	0.5206
DecisionTree (SMOTE)	0.6578	0.6342	0.5459	0.5351
GaussianNB	0.8083	0.6183	0.5321	0.5273
DecisionTree (SMOTEENN)	0.5785	0.5566	0.5224	0.5186
AdaBoost (SMOTE)	0.7825	0.5423	0.5282	0.5227
RandomForest (SMOTEENN)	0.7827	0.5406	0.5218	0.5176
AdaBoost (SMOTEENN)	0.8003	0.5344	0.5291	0.5257
XGBoost (SMOTEENN)	0.7761	0.5280	0.5146	0.5124
XGBoost (SMOTE)	0.7655	0.5141	0.5122	0.5110
RandomForest (SMOTE)	0.7652	0.5125	0.5068	0.5063
AdaBoost	0.8372	0.5000	0.4954	0.4910
DecisionTree	0.8329	0.5000	0.4954	0.4910
RandomForest	0.8349	0.5000	0.4954	0.4910
LogisticRegression	0.8371	0.5000	0.4954	0.4910
XGBoost	0.8380	0.5000	0.4954	0.4910

---

Project Topic

Training a binary classification model to predict strokes in patients

Project Objective

Differentiating high-risk patients for stroke to enable early treatment and prevention through additional testing

Data Description

train_strokes.csv - Clinical information of 43,000 patients

• Imbalanced data with 98.2% of patients not experiencing a stroke and 1.8% having experienced a stroke.

id, gender, age, hypertension, heart_disease, ever_married, work_type, Residence_type, avg_glucose_level, bmi, smoking_status, stroke

Project Summary

Setting stroke experience as the target variable, various models were trained using GridSearchCV

• Resampling methods such as SMOTE and SMOTEENN were used to adjust the imbalanced data distribution

Models: Logistic Regression, Decision Tree Classifier, RandomForest Classifier, XGB Classifier, AdaBoost Classifier, GaussianNB

< Image Source: A Comparison of Undersampling, Oversampling, and SMOTE Methods for Dealing with Imbalanced Classification in Educational Data Mining >

Project Results

• Compared model performance using AUC score, Recall, F1 score, and Precision as metrics.
• Focused on the Recall score as the main objective for the purpose of the model is to predict high-risk stroke patients(True Positives).
• Shown improved results with resampling techniques applied to the models.

Model	AUC Score	Recall	F1 Score	Precision
LogisticRegression (SMOTEENN)	0.8374	0.7565	0.4740	0.5233
LogisticRegression (SMOTE)	0.8356	0.7554	0.4729	0.5230
GaussianNB (SMOTE)	0.8031	0.7384	0.4543	0.5199
GaussianNB (SMOTEENN)	0.8064	0.7365	0.4632	0.5206
DecisionTree (SMOTE)	0.6578	0.6342	0.5459	0.5351
GaussianNB	0.8083	0.6183	0.5321	0.5273
DecisionTree (SMOTEENN)	0.5785	0.5566	0.5224	0.5186
AdaBoost (SMOTE)	0.7825	0.5423	0.5282	0.5227
RandomForest (SMOTEENN)	0.7827	0.5406	0.5218	0.5176
AdaBoost (SMOTEENN)	0.8003	0.5344	0.5291	0.5257
XGBoost (SMOTEENN)	0.7761	0.5280	0.5146	0.5124
XGBoost (SMOTE)	0.7655	0.5141	0.5122	0.5110
RandomForest (SMOTE)	0.7652	0.5125	0.5068	0.5063
AdaBoost	0.8372	0.5000	0.4954	0.4910
DecisionTree	0.8329	0.5000	0.4954	0.4910
RandomForest	0.8349	0.5000	0.4954	0.4910
LogisticRegression	0.8371	0.5000	0.4954	0.4910
XGBoost	0.8380	0.5000	0.4954	0.4910