Small-World Inspired Neural Network (SW-NN) for an Effective and Accurate Diagnosis of Diabetes

As the use of artificial intelligence in medical diagnosis is growing, the need for enhanced efficacy is becoming paramount. By merging the robustness of neural networks with the distinctive attributes of small-world networks (SWN), the possibility arises for even greater levels of accuracy with maintaining good generalizability. This work presents a novel approach for diagnosing diabetes using a Small-World inspired Neural Network (SW-NN). By constructing a SW-NN using Newman-Watts algorithm, we employ its architecture to classify diabetes and compare it to a conventional NN. The SW-NN outperforms the conventional NN in classification accuracy and evaluation metrics as shown by the experimental results. Our accuracy analysis conclusively demonstrates the superiority of the SW-NN, highlighting its potential to improve diabetes diagnosis.

I. Introduction

According to clinical investigations [1], diabetes has been associated with the nervous system, kidney, and cardiovascular diseases, highlighting its significant and widespread health concern. Gestational diabetes, type-1 diabetes, and type-2 diabetes are its widely prevalent forms. Gestational diabetes refers to elevated blood glucose levels in pregnant women. When the immune system itself eliminates the cells present in pancreas that are responsible for insulin production, it is labeled as Type 1 diabetes [2]. When the cells become very less responsive to the insulin, that is the effect of insulin is nullified, this form of diabetes is termed as Type 2 [2]. Due to the accelerating global spread of diabetes attributed to changes in dietary preferences [16], early diabetes detection has become crucial in the medical field.