By 2030, the number of deaths due to cardiovascular disease is expected to increase to 23.3 million [1]. The veins of the heart transport oxygen; if these vessels become blocked or restricted, heart disease or stroke may occur. According to the World Health Organization, approximately 12 million people are affected by heart disease each year, and 80% of those affected die as a result of this condition [2]. The key variables that affect the heart include high cholesterol, high blood pressure, tension, stress, alcohol use, a sedentary lifestyle, obesity, and diabetes. These characteristics can aid in the detection of cardiac disease. Higher blood pressure causes thickening of the arterial walls, resulting in obstruction and an increased mortality rate [3]. The early detection and treatment of cardiac disease can help to prevent and lower patient death rates [4]. Several researchers in the medical field have used various supervised and unsupervised machine learning (ML) algorithms to diagnose and predict cardiac disease. ML approaches aid in the extraction of usable knowledge from large datasets [5].

These algorithms have found widespread use in healthcare, speech recognition, cosmology, and education. Several techniques have been proposed for detecting various patterns in huge datasets [6]. ML is widely used in the healthcare industry to diagnose diseases and make informed decisions. This aids in the classification of patients with significant risk factors. Healthcare businesses collect massive volumes of data, and ML provides several models for fast training and data analysis [7]. These algorithms seek a large search space for optimal solutions by training a dataset. To assess the performance of the models, many measures such as accuracy, sensitivity, specificity, precision, and F1-score may be utilized. The optimum hyperparameters are identified and applied to the ML models by modifying the hyperparameters, which enhances the performance. Robust models can be created by tweaking the hyperparameters. The hyperparameters can be tuned to prevent overfitting or underfitting [4, 8].

This study presents a unique optimization-controlled ML model for cardiovascular disease identification based on patient heart disease data. An artificial neural network (ANN) classifier is used to classify cardiac disease, and the weights of the ANN classifier are optimized using the grey wolf optimization (GWO) method. To make the classification process effective, significant statistical characteristics such as mean, variance, kurtosis, standard deviation, and skewness are retrieved from the input data. Furthermore, the findings of this study reveal a complete description of the proposed model, allowing the efficacy of the proposed categorization model to be confirmed. In addition, a comprehensive analysis is presented based on the performance of traditional classification techniques to assess the benefits of the proposed heart disease strategy.

The major contributions of this research are summarized as follows:

• • A novel artificial intelligence (AI)-based heart disease prediction strategy is designed and developed for advanced healthcare systems.

• • The GWO algorithm is used to enhance the prediction accuracy of the ANN classifier by optimally tuning the weights and parameters.

• • The performance of the ANN model is validated by comparing the proposed and existing methods for the heart disease prediction module.

The remainder of this paper is organized as follows: Section 2 outlines modern heart disease categorization methodologies, as well as the issues that they face. The proposed GWO-based ANN model for heart disease classification is described in Section 3. The outputs of the proposed model are discussed in Section 4, and the study is concluded in Section 5.

The existing techniques for heart disease categorization and the problems associated with the models are discussed in this section, serving as inspiration for the creation of the proposed model.

2.1 Related Works

A literature review of existing methods is presented herein. Abdel-Basset et al. [9] developed an Internet of Things (IoT)-based approach for identifying and monitoring cardiac patients. The purpose of the healthcare model is to increase diagnostic precision. The precise answer to the model could lead to a reduction in mortality and treatment costs. Particle swarm optimization (PSO) and rough sets (RS) with transductive support vector machines (TSVM) was created by Thiyagaraj and Suseendran [10] for heart disease detection. To reduce data redundancy, this strategy increases data integrity. Data are normalized using the zero-score method. PSO is then used to select the best subset of characteristics, reducing the computing cost and improving the prediction performance. Finally, a radial basis function TSVM classifier is used for heart disease prediction. Shah et al. [11] developed an automated diagnostic approach for cardiac disease. Using the advantages of feature selection and extraction models, this technique assesses relevant feature subsets. Magesh and Swarnalatha [12] developed a model for detecting heart disease using Cleveland cardiac samples. Cluster-based decision tree (DT) learning is used to identify cardiac problems. The intended label distribution is used to split the original set and the potential class is determined using the elevated distribution samples. Entropy is employed to determine the characteristics of each class set for heart disease diagnosis. Nilashi et al. [13] used ML algorithms to develop a prediction approach for heart disease detection. This system combines unsupervised and supervised learning to identify cardiac problems. For missing value assertion, the approach uses a self-organizing map, fuzzy SVM, and principal component analysis (PCA). Furthermore, incremental PCA and the fuzzy SVM are designed for incremental data learning to reduce the computation time in disease prediction.

2.2 Problem Statement

The challenges associated with the existing methods for heart disease classification are as follows:

• • The problem with heart disease data is that they require feature selection, which is complicated by the difference in samples and the absence of feature magnitudes. Furthermore, the explanation of cardiac echo images is difficult because of the lack of defined criteria for accurately inferring the data [1].

• • Large amounts of original data have been created during the digital age. Gathering useful data is a significant challenge and is an ideal area for exploration. Several ambiguities are present in the information systems that are used for disease diagnosis, making them difficult to handle [4].

• • For the diagnosis of cardiac disease, patient data include test results, video clips, and demographic information. Mining valuable information from large amounts of data is time consuming [9].

2.3 Motivation

Heart disease is the most common cause of death worldwide. India is in the midst of an industrial revolution that has resulted in lifestyle changes that contribute to heart disease. Despite the information contained in medical data, clinicians diagnose diseases based on their experience and individual opinions, which can lead to inaccurate diagnoses and high diagnostic expenses, thereby affecting the quality of services offered by hospitals to patients [14]. Medical data mining aims to discover useful information for accurate diagnoses. Disease therapy is difficult and time consuming in healthcare settings. The identification of cardiac diseases resulting from several circumstances is considered a multi-layered problem. Hence, it is necessary to develop an automated model for the accurate prediction of heart disease.

This study presents a method for predicting heart disease using an ML model and patient data. The data are analyzed so that the cardiac problems of individuals can be anticipated automatically. Appropriate tuning of the ANN parameters aids the system in improving its effectiveness in predicting cardiac diseases. Figure 1 provides a diagrammatic representation of the proposed model for heart disease prediction. ANNs are computational models inspired by the neural architecture of the human brain, and consist of interconnected layers of nodes or neurons. Each connection has an associated weight, and ANNs can learn complex patterns and relationships in the data through a training process that typically uses algorithms such as backpropagation. They are widely used for tasks such as classification, regression, and pattern recognition owing to their ability to model nonlinear relationships and handle high-dimensional data. Integrating ANNs with optimization algorithms such as the GWO algorithm can enhance their performance. GWO is a nature-inspired metaheuristic that mimics the social hierarchy and hunting behavior of grey wolves. It involves four types of wolves–alpha, beta, delta, and omega–representing the hierarchy within the pack. The algorithm iteratively updates the positions of the wolves based on the best solutions found thus far, guided by the alpha, beta, and delta wolves. Combining ANNs with GWO leverages the strengths of both, using GWO to optimize the weights and biases of the ANN, thereby improving the ability of the network to find optimal or near-optimal solutions to various complex problems. This hybrid approach enhances the learning capabilities and overall performance of the ANN, making it effective for a wide range of applications.

3.1 Input Heart Disease Data

Consider the input heart disease medical dataset I with various attributes, which is expressed as

where I_A,B denotes the Bth attribute A in the data, p specifies the total data, and q is the total attributes in each data. The dimension of the database is indicated as [p × q].

3.2 Pre-processing of Heart Disease Data

The purpose of pre-processing is to make the input data processing more efficient. Pre-processing is an important phase that prepares the input data for the prediction process. In this step, medical records are converted into diagnostic values as part of the data pre-processing.

3.3 Feature Extraction Process

The features are obtained from the heart disease data once the pre-processing stage is completed. Statistical features, such as mean, standard deviation, variance, kurtosis, and skewness are retrieved from the pre-processed data. With the aid of statistical features, enormous amounts of data can be analyzed to find common patterns and trends, and then transformed into useful information. The extracted features are described in detail below.

a) Mean: The mean is a statistical characteristic that estimates the average of all data instances in proportion to the total occurrences from the data, and is expressed as

where Q is the total number of times that the data occur and K_a denotes the ath data point.

b) Variance: The average of the squared variances between the individual data and mean is the factor variance and is represented as

The variance measure tracks minute deviations in data over time, which helps to enhance the diagnosis process considerably.

c) Standard deviation: The measurement of the standard deviation is the root of the variance, which is the evaluation of widely scattered data responsible for evaluating each occurrence of heart disease data concerning the mean. The standard deviation equation is as follows:

d) Skewness: The skewness is defined as the rate of symmetry or lack of symmetry, and is defined as the rate of the third central moment. The skewness of a dataset with perfect symmetry is always zero, and the skewness is made zero by the normal distribution. The formula for skewness is as follows:

e) Kurtosis: The kurtosis is defined as the value of the distributed weight of the ends corresponding to a continuous distribution, and is equal to zero in the case of a Gaussian distribution. The kurtosis formula is expressed as follows:

where f_SD is the standard deviation, F_M is the mean, and K_a indicates the ath measure of K.

3.4 Establishment of the Feature Vector

The feature vector is created using the input data, which include all vital statistics of the patient. These features are combined to form a feature vector, which is expressed as

The dimension of the feature vector is [1×5], and the feature vector is used as the input by the ANN to categorize the input as normal or abnormal.

3.5 ANN Classifier in Heart Disease Classification

Heart disease can be classified using ML algorithms, which detect conditions without the need for human interaction. The results obtained with ANNs offer many benefits in a range of applications compared with other ML approaches. An ANN is a mathematical model based on a biological neural network composed of a group of artificial neurons that calculate information using a connection mechanism. ANNs are computer models with many interconnections and simple, highly parallel processors. The primary premise behind the development of the ANN is that it is based on human characteristics and has neuron layers for analyzing the data that it receives as input. Multiplying the ANN input by weights [15] represents the data flow. Mathematical functions are used to assess the weights, resulting in the activation function of the neuron. The appropriate weight-tuning strategy of the GWO algorithm allows the output of the ANN to be controlled as desired. The construction of the ANN is illustrated in Figure 2.

The output of the ANN can be mathematically expressed as

where P is the activation function, r_n is the input, and G_en is the weight measure. An ANN classifier is used to divide the input as normal or diseased, and the benefit of employing an ANN [15] is that the classifier automatically executes the classification operation without requiring human intervention. The classification accuracy depends on the accurate tuning of the hyperparameters of the ANN classifier by the GWO algorithm.

The GWO optimization technique is a metaheuristic based on the hierarchy-based hunting characteristics of the Canidae family [16]. The aspects of the optimization method focused on hunting behaviors indicate the dynamic nature of addressing convergence concerns. TheGWO algorithm solves this problem by leveraging the dynamic features of the Canidae population. This improves the ability of the algorithm to achieve better global optimal convergence than other optimization techniques that frequently converge to local optimal solutions, with better stability between the exploitation and exploration phases. In the presence of grey wolves, the hierarchy-based hunting phenomenon exhibits a hunting hierarchy, including a leading grey wolf, a representative grey wolf, an obeyer grey wolf, and a scapegoat grey wolf.

As stated previously, the hierarchy-based hunting experience of the population is supported by hierarchical levels, with the leading grey wolf at the top of the hierarchy and in charge of devising solutions for resting, hunting, and other activities. During hunting, the leader grey wolf teaches the other grey wolves, despite the fact that the leading grey wolf is not the strongest and does not have any mates within the group, but remains the best grey wolf in the group. At the second-order level, the representative grey wolf directs the leading grey wolf by transferring the decision to the subordinate grey wolves in the search area. The scapegoat grey wolves remain at the bottom of the food chain, guarding the entire force. The obeyer grey wolves lie directly above the scapegoat grey wolf and mostly follow the orders of the leading grey wolves. The proposed GWO algorithm comprises four primary phases, as follows.

a) Social tracking hierarchy: The leading, representative, obeyer, and scapegoat grey wolves are pursued by the grey wolf hierarchy, with the leading grey wolf being the strongest among all grey wolves. The representative and obeyer grey wolves compete for second and third place, respectively, and are considered as the second- and third-best grey wolves, whereas the remaining grey wolves in strength are regarded as scapegoat grey wolves.

b) Surrounding phase: The grey wolf surrounds the prey in this phase based on the needs and convenience of the grey wolf. In the surrounding phase, the grey wolf is modeled as

where Ebcs+1 indicates the position of the bth grey wolf at the cth coordinate axis and represents the updated location. Let ERs represent the location of the prey, C represent the space between the prey and grey wolf, and U be the coefficient vector. The distance vector is formulated as

where E_s represents the location of the grey wolf at time s and W₁ represents the coefficient vector. The coefficient vectors U and W₁ are mathematically modeled as

where g progressively decreases from 2 to 0 as the number of iterations increases, and m₁ and m₂ denote random vectors ranging from 0 to 1.

c) Grey wolf hunting phase: Grey wolves can recognize prey, and the hunting process is carried out with the help of the leader grey wolf, representative grey wolf, and obeyer grey wolf. The general characteristics of grey wolves are as follows:

where C_J, C_K, and C_L represent the current route of the best grey wolf at the time of hunting. W₁, W₂, and W₃ symbolize the random limits of the grey wolf in balancing the exploration and exploitation phases. The distance vectors are expressed as

where U₁, U₂, and U₃ indicate the coefficient vectors and E_j is the best solution at time t. U₁, U₂, and U₃ represent the tunable parameters in the optimization process. In general, the hunting process among grey wolves is based on the first three best solutions, according to the roles of the leader, representative, and obeyer grey wolves [18].

where E₁ indicates the position of the leader grey wolf, E₂ is the position of the representative grey wolf, and E₃ is the position of the obeyer grey wolf of the optimization procedure, with reference to the three best hunting locations. The hunting is led by the space between the grey wolf and prey, as well as the tunable coefficient parameters, as shown in the above modeling. As a result,

Eq. (20) is the updated formulation of the GWO optimization method, which is based on the constraints imposed by the grey wolf hierarchy-based hunting characteristics. Using Eq. (20), the GWO optimization method improves the accomplishment of the global best solution.

d) Social dominance stage: The grey wolves only probe the prey inside the search region during this phase and pick the ideal local sites to attack the prey.

e) Exploration stage: The grey wolves, including the leader, representative, obeyer, and scapegoat, investigate the environment outside the resolution space in this phase and arrive at the best possible conclusion. The pseudocode of the GWO algorithm is presented in Table 1.

This section discusses the outcomes of the GWO-ANN prediction module, and a comparative analysis demonstrates the performance of the GWO-ANN in cardiovascular disease prediction.

5.1 Experimental Setup

The study was conducted using Python, which was installed on a Windows 10 64-bit computer with 16 GB of RAM.

5.2 Database Description

a) Heart disease database: The heart disease database was used as the dataset in the proposed cardiovascular disease prediction module. Of the 76 attributes in this database, only 14 are frequently assessed for research purposes. Using these data, an experimental study was conducted to determine the presence or absence of cardiac disorders [17].

b) Statlog + Cleveland + Hungary database: This dataset was created by integrating several datasets that were previously accessible individually but had never been merged. The Heart Disease UCI and Statlog + Cleveland + Hungary datasets, both available on Kaggle, are essential resources for predicting heart disease using ML. The Heart Disease UCI dataset contains 303 instances with 14 attributes, including age, sex, chest pain type, resting blood pressure, serum cholesterol, fasting blood sugar, resting electrocardiographic results, maximum heart rate achieved, exercise-induced angina, ST depression induced by exercise, slope of the peak exercise ST segment, number of major vessels colored by fluoroscopy, thalassemia, and a target variable indicating the presence of heart disease. This dataset allows for comprehensive analysis and feature engineering, making it useful for developing predictive models. The Statlog + Cleveland + Hungary dataset combines data from three major heart disease studies, and offers a robust dataset with 918 instances and the same 14 attributes. Merging records from the Cleveland, Statlog, and Hungary datasets provides a more extensive and diverse set of patient information, enhancing the generalizability and robustness of heart disease prediction models. This dataset is particularly valuable for analyzing trends and creating models that can predict heart disease across different populations.

These databases were integrated based on 11 similar traits, resulting in the world’s largest heart disease dataset [18].

5.3 Evaluation Metrics

a) Accuracy: In optimal medical data categorization, the accuracy is defined as the scale on which an evaluated value is close to its actual measure.

b) Sensitivity: The ratio of positives accurately detected by the classifier is referred to as the sensitivity.

c) Specificity: The ratio of properly detected negatives by the classifier is known as the specificity.

d) F1-score: The F1-score, which is the average of the precision and recall, is a measure of a model’s accuracy.

5.4 Comparative Methods

To validate the efficacy of the proposed technique, methods including the multilayer perceptron (MLP) classifier, SVM classifier, random forest (RF), AdaBoost classifier [19–22], and ANN classifier [8] were compared with the proposed GWO-ANN classifier in terms of the accuracy, sensitivity, specificity, and F1-score.

5.5 Comparative Analysis

This section presents a comparative evaluation of the approaches used in the classification of heart disease using conventional techniques and the proposed model (Table 2).

This study has presented an ML-based heart disease prediction module for the diagnosis of cardiovascular diseases by analyzing heart disease data. The heart disease data were initially preprocessed to eliminate artifacts. Significant statistical features such as the mean, standard deviation, variance, kurtosis, and skewness were then extracted from the data, through which a feature vector was developed that acted as the input to the proposed ANN classifier. The performance of the ANN was enhanced by optimally tuning the weights using the GWO algorithm, which is derived by considering the hierarchy-based hunting characteristic features of grey wolves. The efficiency of the proposed model was verified using performance measures including the accuracy, sensitivity, specificity, and F1-score, with results of 92.9245%, 94.7469%, 90.9215%, and 96.0551%, respectively. In future, the accuracy of the model will be further enhanced by optimally tuning the ANN classifier using hybrid metaheuristic optimization algorithms. Furthermore, this method can be tested for real-time applications in the medical field.

Table 1. Pseudocode of GWO algorithm.

Table 2. Comparison of heart disease classification techniques.