Article Preview
Top1. Introduction
Due to advancement in health care system, life expectancy of human is increased tremendously in last two decades. But, several challenges are associated with healthcare systems. These challenges are lack of medical information, diagnostic errors, inadequate data, irrelevant features, data threaten etc. In literature, several expert systems, decision diagnostic system, electronic health record system, prediction system are presented (Doi, 2007; Meystre et al., 2008; Rangayyan et al., 2007). These systems considerably help the physicians for disease diagnosis. The term disease diagnosis refers to determine the disease through symptoms and it can be formulated as classification of medical data for decision making. Such decision making systems requires higher computing power to process the large medical dataset. The decision capabilities of these systems depend on amount of training data. The aim of these systems is to minimize the physician errors. Such systems can enhance the decision capabilities of physician as well as improve the diagnostic accuracy. It is noticed that many computer-aided diagnostic tools are presented in literature to help the physicians. Large number of machine learning techniques is incorporated in diagnostic tools to improve the prediction rate (Kavakiotis et al., 2017; Nahar et al., 2013; Nilashi et al., 2017; Shickel et al., 2017). It is seen that all features are not equally important for decision making process. The irrelevant features can also affect the prediction accuracy of the algorithm. Hence, several researchers also focus on the identification of relevant features for disease diagnosis and prediction. Several features selection algorithms are also presented in literature (Akay, 2009; Inbarani et al., 2014; Lin & Hsieh, 2015; Subanya & Rajalaxmi, 2014). Features selection can be defined as identification of most relevant set of features from the medical datasets. The aim of feature selection techniques is to reduce computational cost and also improve the accuracy rate of diagnostic process. These techniques are also integrated with machine learning methods and can be either supervised or unsupervised. The aim of these feature selection techniques is to compute a weight function for each feature of the medical dataset. The weight function is used to determine the relevant features form datasets. Some of feature selection techniques to improve predictive accuracy are listed as. A particle swarm optimization (PSO) and ABC based feature selection algorithm to determine the optimal set of features is reported in (Lin & Hsieh, 2015). To determine the optimal set of features for support vector machine (SVM), Subanya et al. (2014) applied ABC based feature reduction algorithm on heart dataset. Akay applied F-score technique to determine optimal set of features for breast cancer dataset (Akay, 2009). Inbarani et al. (2014) adopted rough set based approach to remove redundant features from medical datasets. Further, this approach is integrated with PSO for improving the classification accuracy. Hence, feature selection is an important issue for building an efficient and effective diagnostic model. Further, it is also limit the number of input features in the diagnostic model in order to produce good predictive results (Akay, 2009). Hence, in this work, an attempt is made to select the optimum features for deep neural network model to improve the diagnostic accuracy. To address the same, ABC based feature selection method is designed to determine the optimum set of features. Further, this feature selection method is integrated with DNN technique. Finally, a diagnostic model is developed using ABC based feature selection method and DNN technique, called ABC-DNN diagnostic model. The performance of proposed ABC-DNN diagnostic model is evaluated using Pima Indian Diabetes disease dataset. It is noticed that proposed ABC-DNN diagnostic model provides better results than DNN technique. The main contributions of this study are summarized as follows:
- 1.
To develop an efficient and effective diagnostic model for improving the diagnostic accuracy of diabetes disease.
- 2.
ABC based feature selection method is designed to select the relevant feature for the diagnosis of diabetes disease.
- 3.
DNN technique is adopted for the prediction of diabetes disease using reduced feature set.
- 4.
This paper explore the impact of DNN technique on the accuracy issue of diabetes disease using feature selection and not using the feature selection technique. This paper also investigates the impact of feature selection technique on a diagnostic model either the performance of diagnostic model is improved or not improved. Hence, this paper also described the role of feature selection technique in the field of healthcare.