Article Preview
Top1. Introduction
Information mining is a powerful way to drawing meaningful information from datasets containing massive embedded data. Data mining may be fruitfully exploited in hospitals where a huge volume of data exists. These hospital datasets often need to be clustered, or even further classified to gain a meaningful analysis of the underlying data. Soft computing techniques such as pattern recognition (PR) and machine learning (ML) have been used for identifying statistical parameters from the dataset. A World Health Organization (WHO) report claims that almost 422 million people worldwide are suffering from both TYPE-1 and TYPE-2 diabetes (https://www.who.int/health-topics/diabetes). Also, as released by the WHO, India’s profile shows that 46% of the total population have prevailing diabetes and related risk factors.
Diabetes Mellitus (DM) is mainly sorted as three types: (a) diabetes mellitus or DM, (b) insulin resistance, and (c) the third one is gestational diabetes generally found amid pregnant ladies. DM is generally caused by high blood glucose levels and is a typical disease that influences those individuals who have imbalance in blood glucose levels, especially for pregnant women. In fact, past research has demonstrated that pregnant women with diabetes are increasingly inclined to have newborns with birth defects than those without diabetes. Kanguru, et al. (2014) notes that, in such situation, the child may be influenced by prevailing illness conditions, for example, coronary illness and spina bifida. The beginning stage for living admirably with DM is an early diagnosis. Hence, the primary aim of our work is to predict diabetic or non-diabetic using ML and deep learning algorithms. A major study of the classification systems gives high accuracy with high handling time, though few strategies have yielded low precision even with enormous dataset. Along these lines, our work aims for high accuracy and less processing time with immense dataset.
Figure 1. Schema of a Machine Learning (ML) Model
Figure 1
represents a schema of the ML model utilized in our work. Nowadays, many efficient analysis techniques are available at affordable cost. Data analytic approaches improve the disease detection accuracy in modern hospitals. DM, when detected early, can avoid serious complications and may also be managed via a healthy diet. The key objective here is to build a prediction system by combining the usefulness of ML and Learning Model algorithms.
Following the introduction, Section 2 reviews the extant research in diabetes detection using ML and deep learning approaches. Section 3 offers insights into our proposed novel approach with details on the PIMA dataset vis-à-vis details on the correlation-based feature selection (CFS) subset evaluation algorithm. Section 4 shifts focus into the data visualization with details on the key analytic approaches employed, namely, enhanced support vector machine (ESVM) and deep neural network (DNN). Section 5 then presents the results and reflections. Finally, in Section 6, we offer concluding remarks with insights into future research directions.
TopNumerous works have exploited ML, clustering, classification, information mining and learning applicable for diabetes discovery systems. Here, we survey the more prominent works in the extant literature with their concise propositions.