Back to Projects

Diabetes Prediction

Stacking ensemble combining 4 base learners with a meta-model.

Stackingscikit-learnCross-ValidationIEEE

Ensemble Architecture

Stacking Architecture

2-Layer Ensemble
Logistic78%Random80%Gradient79%SVM77%Meta-LearnerLayer 1Layer 2

Accuracy Comparison

+2.68% improvement
75%80%85%78%Logistic80%RF79%GBoost77%SVM82.68%Stack

Feature Importance

Glucose
26%
BMI
18%
Age
15%
Pregnancies
12%
Blood Pressure
10%
Insulin
8%
Skin Thickness
6%
Pedigree
5%

Training Pipeline

PreprocessScale + Impute
Base Models4 Classifiers
Meta-LearnerLogistic Reg
Evaluate5-Fold CV
82.68%
Accuracy
4
Base Models
5-Fold
Cross-Validation
8
Features
Published in IEEE AIMV-21