Project goals

• Use 3 similar cleaned datasets to find the features that best predict diabetes in patients.

Project description

• I've taken 2 datasets from kaggle that were cleaned for modeling to determine diabetic patients.

The Plan

Initial hypotheses

• What features will have the greatest weight on determining diabetic patients?
• Using the encoded target variable diabetit_012, how will each set perform on models and will this effect how future questionaires should be formatted?

Acquire:

I acquired 2 datasets from Kaggle as csv downloads and saved them in my local folder from this site:

• https://www.kaggle.com/datasets/alexteboul/diabetes-health-indicators-dataset/data?select=diabetes_012_health_indicators_BRFSS2015.csv

• diabetes _ 012 _ health _ indicators _ BRFSS2015.csv

• diabetes _ binary _ health _ indicators _ BRFSS2015.csv

Prepare:

• Lowercased the column names
• Created a third dataframe with equal weight of the target variables on diabetes_012 0=No Diabetes / 1=Prediabetic / 2=Diabetic
  • 13983 Rows / 22 Columns

Explore:

• Explored on the first data set
• How does the percentage of diabetics/prediabetics and the features appear in stacked percentages?
• What do the features look like based on Age?
• What features is diabetes_012 dependent on?

Modeling:

• Modeling was done for all 3 sets.
• Use drivers in explore to build predictive models of different types
• Evaluate models on train and validate data
• Select the best model based on accuracy
• Evaluate the test data

Data dictionary:

Feature	Definition
Diabetes_012	Ever Told you have diabetes 0=No/OnlyDuringPrgnancy, 1=Prediabetic, 2=Diabetic
HighBP	Have you ever been told you have High Blood Pressure, 0=No, 1=Yes
HighChol	Have you ever been told you have had High Cholesterol, 0=No, 1=Yes
CholCheck	Cholesterol Check in last 5 years?, 0=No, 1=Yes
BMI	Body Mass Index Number
Smoker	Smoke at least 100 cigarettes,etc in your entire life?, 0=No, 1=Yes
Stroke	Ever had a stroke, 0=No, 1=Yes
HeartDiseaseorAttack	Respondents that have ever reported having coronary heart disease or myocardial infarction? 0=No, 1=Yes
PhysActivity	Adults who reported doing physical activity or exercise during the past 30 days other than their regular job, 0=No, 1=Yes
Fruits	Consume Fruit 1 or more times per day, 0=No, 1=Yes
Veggies	Consume Vegetables 1 or more times per day, 0=No, 1=Yes
HvyAlcoholConsump	Heavy drinkers (adult men having more than 14 drinks per week and adult women having more than 7 drinks per week), 0=No, 1=Yes
AnyHealthcare	Do you have any kind of health care coverage, including health insurance, prepaid plans such as HMOs, or government plans such as Medicare, or Indian Health Service? 0=No, 1=Yes
NoDocbcCost	Was there a time in the past 12 months when you needed to see a doctor but could not because of cost? 0=No, 1=Yes
GenHlth	Would you say that in general your health is: 1 to 5 scale, 1 is excellent, 5 is poor
MentHlth	Now thinking about your mental health, which includes stress, depression, and problems with emotions, for how many days during the past 30 days was your mental health not good? Scale 1 - 30
PhysHlth	Now thinking about your physical health, which includes physical illness and injury, for how many days during the past 30 days was your physical health not good? Scale 1 - 30
DiffWalk	Do you have serious difficulty walking or climbing stairs? 0=No, 1=Yes
Sex	Indicate sex of respondent. 0= Female, 1= Male
Age	Fourteen-level age category,1 is 18-24, all the way to 80 and Older, 5 yr increments
Education	What is the highest grade or year of school you completed? 1 - 6, 1 is never attended school or kindergarten, 6 is college 4 yrs or more
Income	Is your annual household income from all sources: (If respondent refuses at any income level, code "Refused.") 1-8, 1 is < 10k, 8 is > 75k

How to Reproduce

• Clone this repo
• Acquire data from Kaggle in link above in Acquire and save csv's to local folder.
• Run Notebook

Key findings

• It seems highbp, highchol, smoker, stroke, and heardiseaseorattack greatly increases the percentage of diabetics and prediabetic patients as they get older.
• Having a lower income/education level puts younger patients at higher risk of diabetes
• BMI visually shows as bmi increases, so does the amount of patients with diabetes.

Takeaways and Conclusions

• Logistic Regression Models worked best on all 3 datasets
• Highbp, highchol, bmi, heartdiseaseorattack are big indicators for diabetes.
• It is very difficult to predict diabetes off of a questionairre, however, looking at the way the prediabetic patients answered their questions is a very good way to start predicting diabetes.

Recommendations

• Have more questions tailored to finding more information of pre-diabetic patients.