PREDICTING DIABETES USING DEEP LEARNING TECHNIQUES: A STUDY ON THE PIMA DATASET

Abstract

Diabetes is one of the key reasons of growing death rates around the world. Diabetes is a medical
condition that arises from chronic issues that influence carbohydrate metabolism and raise blood glucose
levels. Scientific research is needed to diagnose diabetes early for prevention and treatment due to the
growing rates of the disease. Researchers have recognized the value of classification models for disease
prediction developed using machine learning and deep learning techniques. This study explores the
efficacy of deep learning techniques Convolutional Neural Network (CNN), Long Short-Term Memory
(LSTM), and Multi-Layer Perceptron (MLP)—in forecasting diabetes using the Pima dataset.
Preprocessing steps encompassed normalization, handling missing values, and outlier removal. The
models were trained and evaluated, yielding noteworthy performance metrics. The CNN exhibited the
highest accuracy of 0.77, while achieving precision, recall, and ROC-AUC scores of 0.69, 0.67, and 0.83,
respectively. The LSTM and MLP models also demonstrated competitive results, achieving accuracies of
0.75 with similar precision, recall, and ROC-AUC values around 0.64-0.67 and 0.80-0.82, respectively.
These findings highlight the potential of deep learning methodologies for predictive diabetes analysis and
emphasize the significance of proper preprocessing techniques in enhancing model performance.