Predictive Modelling of Diabetes

Diabetes has become a leading cause for concern amongst Australia's population, with over 1.3 Million people confirmed cases and 500,000 more estimated undiagnosed type 2 cases.¹

The scale of this issue is compounded by the fact that on average each of these people typically have a family member or carer who's life is directly effected in a support capacity and that diabetes is the 7th leading cause of death of Australians.²

It is from this perspective that we aim to develop a predictive model for diabetes risk to improve the ability for medical professionals to make timely detections and interventions as well as to give users agency in preventing this disease.

Data Analytics

To analyse this data, as well as to train machine learning classification models, a number of python libraries have been incorporated into the project:

• Data Analytics / Transformation:
  • Pandas
  • Numpy
• Visualization:
  • MatPlotLib
  • Seaborn
• Machine Learning:
  • Scikit-Learn
  • Tensorflow / Keras
  • KerasTuner

Dataset 1

https://www.kaggle.com/datasets/iammustafatz/diabetes-prediction-dataset
The original poster is coy³ about the source of the data but it is still one of the most popular datasets across Kaggle.

This Dataset consists of 100001 records which after cleaning, and selction for those aged 55 and over, is reduced to 30995.

The columns are as follows

Name	Description	Type
gender	Gender of the patient	Nominal: [Male, Female, Other]
age	Age of the patient	Discrete: [Years, 55 ➡ 80]
hypertension	History of hypertension	Nominal: [0: false, 1: true]
heart_disease	Patient history of heart disease	Nominal: [0: false, 1: true]
smoking_history	Patients Smoking History	Nominal: ['never', 'current', 'No Info', 'former', 'not current', 'ever']
bmi	Body-Mass Index: calculated by body-mass / height^2	Continuous
HbA1c_level	Average Blood Glucose over the last 3 months	Continuous: [A1c%]
blood_glucose_level	Blood glucose at time of recording	Discrete: [mg/dL]
diabetes	Whether the patient has been diagnosed with diabetes	Nominal: [0: false, 1: true]

Dataset 2

https://archive.ics.uci.edu/dataset/529/early+stage+diabetes+risk+prediction+dataset

This dataset is provided in the UC Irvine ML repo and was collected using questionnaires from the patients of the Syhlet Diabetes Hospital in Bangladesh.⁴

The Dataset contains 17 columns and 520 records

Name	Description	Type
Age	Age of the patient	Discrete: [Years]
Gender	Gender of the patient	Nominal: [Male, Female]
Polyuria	Excessive Urine	Nominal: [No, Yes]
Polydipsia	Excessive Thirst	Nominal: [No, Yes]
Sudden Weight Loss		Nominal: [No, Yes]
Weakness		Nominal: [No, Yes]
Polyphagia	Extreme Hunger	Nominal: [No, Yes]
Genital Thrush		Nominal: [No, Yes]
Visual Blurring		Nominal: [No, Yes]
Itching		Nominal: [No, Yes]
Irritability		Nominal: [No, Yes]
Delayed Healing		Nominal: [No, Yes]
Partial Paresis	Muscular weakness / impairment	Nominal: [No, Yes]
Muscle Stiffness		Nominal: [No, Yes]
Alopecia	Hair Loss	Nominal: [No, Yes]
Obesity		Nominal: [No, Yes]
Class	Presense of Diabetes	Nominal: [Negative, Positive]

Approach

Please discuss your analytics workflow / methodology here

Footnotes

1. https://www.diabetesaustralia.com.au/about-diabetes/diabetes-in-australia/ ↩

2. https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-may-2024 ↩

3. https://www.kaggle.com/datasets/iammustafatz/diabetes-prediction-dataset/discussion/406676#2282358 ↩

4. Islam, M. M. Faniqul et al. “Likelihood Prediction of Diabetes at Early Stage Using Data Mining Techniques.” Computer Vision and Machine Intelligence in Medical Image Analysis (2019): n. pag. ↩