Developed an end-to-end Artificial Neural Network (ANN) pipeline to predict diabetes risk using the CDC BRFSS 2015 dataset. The project focused on model training under class imbalance and evaluation of predictive robustness.

Medical datasets often exhibit significant class imbalance, leading to biased models that over-predict majority classes.

The goal was to build and evaluate a neural network while explicitly analysing the impact of imbalance on training dynamics and predictive performance.

Data preprocessing and feature normalization

ANN implemented using TensorFlow and Keras

Binary classification setup

Model training and validation using imbalanced dataset

Performance evaluation across precision, recall, and F1 metrics

Analysis of decision threshold effects on minority-class recall

The Behavioral Risk Factor Surveillance System (BRFSS) is a health-related telephone survey that is collected annually by the CDC. This original dataset contains responses from 441,455 individuals and has 330 features. The dataset used in this Case Study is a dataset of 253,680 survey responses to the CDC's BRFSS2015 with 21 feature variables. The target variable Diabetes_012 has 3 classes:

0 is for no diabetes or only during pregnancy

1 is for prediabetes

2 is for diabetes.

There is class imbalance in this dataset. The dataset was retrieved from: https://www.kaggle.com/datasets/alexteboul/diabetes-health-indicators-dataset

Built full pipeline in Jupyter Notebook

Explored imbalance effects on gradient updates and convergence

Compared training behavior across multiple hyperparameter settings

Documented trade-offs between recall and false positives

TensorFlow, Keras, Python, Jupyter Notebook