The document discusses predicting the onset of diabetes among women aged 21 and older using binary classification models. It analyzes medical data from 691 training and 77 test patients using eight features to train two models - a boosted decision tree and decision jungle. Both models achieved over 76% accuracy in predicting diabetes onset. The document concludes the models can help identify at-risk patients for timely treatment intervention and help healthcare systems prevent early deaths and health risks through targeted prevention strategies.

1. Page 1
Diabetes Prediction
21/11/16

2. Page 2
BACKGROUND
Diabetes is a major
public health
problem in the world
The total number of
diabetic patients is
expected to soar to
366 million in 2030
1.7 million people 20
years and older were
diagnosed with
diabetes in 2012
alone!
Detailed information
about the prevalence
and demographics of
diabetes are important
for health care
providers to take
suitable approaches

3. Page 3
PROBLEM STATEMENT
To predict the onset of diabetes amongst women
aged at least 21 using binary classification and
compare the results
The model can be used by the endocrinologists,
dietitians, ophthalmologists and podiatrists to
predict if or if not the patient is likely to suffer from
diabetes, if yes, how intense it could be.
The dataset consists of 8 features comprising the
medical details of the patients that are useful in
determining the health condition of the patient.

4. Page 4
MODELING
The train data consists of 691 entries
and the test data has 77 entries
This is a binary classification problem
where 1 indicates that the patient is
likely to suffer from diabetes and 0
indicates that the patient may not
suffer
Two Class Boosted Decision Tree and
Two Class Decision Jungle have been
used to train the model on the 8
features, and the results are
compared
No. of Pregnancies Age Skin Thickness
Blood Pressure BMI
Diabetes Pedigree
Function
Glucose Conc. Insulin Level
The 8 Important Features: Clinically non-invasive & biomarkers

5. Page 5
RESULTS
The results are obtained as follows:
Two Class Boosted Decision Tree Two Class Decision Jungle
Accuracy 76.8 77.2
Specificity 81.2 80.4
Sensitivity 63.8 61.2
AUC 82.8 83.2
Kappa 48.3 48.4
Using a combination of various Clinical non-invasive and biomarkers, Predictive Modeling techniques can thus:
Identify people at high risk of developing diabetes & provide timely intervention in its
treatment to the young, especially women
Healthcare systems can help prevent early deaths and lower health risks by engaging
such predictive techniques
Targeted prevention strategies can be planned to ensure proper efforts to be taken for
prevention in individuals at high risk