Diabetes detection using
machine learning.
By
Namakhanyu Williams.
Meregulwa Allan.
Ategeka Alex.
Project overview.
• Objective: Develop a predictive model to classify individuals as
diabetic or non-diabetic based on clinical features.
• Scope: Focus on binary classification using supervised learning,
with emphasis on accuracy and interpretability for healthcare
applications.
• Key Components: Data preprocessing, model training, ensemble
techniques, and evaluation.
• Expected Outcome: A robust model achieving over 90%
accuracy for early diabetes screening.
MOTIVATION
• Diabetes mellitus, is an emerging public health challenge.
• There is a high burden of undiagnosed cases, late
presentations, and complications like kidney failure, heart
disease, stroke, and amputations.
• Early prediction can enable timely interventions, reducing
mortality and healthcare burdens.
• This project leverages ML to address this challenge.
Dataset overview.
Source:[Link]
datasets/Bena345/diabetes-readmission
(diabetes_prediction_dataset.csv)
Size: 100,000 records
Features: 9 attributes including demographic, clinical, and
behavioral data.
Dataset description.
Feature Type Description
gender Categorical Male/Female
age Numeric Patient age
hypertension Binary (0/1) History of hypertension
heart_disease Binary (0/1) History of heart disease
smoking_history Categorical never/current/former/etc.
bmi Numeric Body Mass Index
HbA1c_level Numeric Average blood sugar (3 months)
blood_glucose_level Numeric Current blood sugar level
diabetes Binary (0/1) Target variable
Data cleaning steps.
• Handle missing values (e.g., smoking_history has "No
Info").
• Check for duplicates and inconsistencies.
• Validate ranges (e.g., bmi, age).
• Encode categorical variables
(e.g., gender, smoking_history)
Proposed Algorithms.
Random Forest: this algorithm is suitable because it is;
Robust ensemble method
Less prone to overfitting than single trees
Good interpretability with feature importance
Reliable baseline performance
Expected outcome.
• A comparative analysis of model performance.
• Identification of key risk factors for diabetes.
• A deployable prototype for diabetes risk assessment.
Challenges and limitations
• Challenges: Data imbalance, missing values (zeros).
Project timeline.
•Week 1-2: Data collection, cleaning, EDA.
•
•Week 3-4: Model implementation and tuning.
•Week 5: Ensemble building and evaluation.
•Week 6: Results analysis, report writing.
•Total Duration: 6-8 weeks (assuming part-time).
conclusion
• This project proposes an accurate ensemble model for
diabetes prediction, addressing a critical health issue.