Heart Disease Detection: A Comprehensive Analysis of Machine Learning, Ensemble Learning, and Deep Learning Algorithms

Figure 1The number of deaths in the UK (England, Wales, Scotland, and Northern Ireland) from 1961 to 2020 caused by all heart and circulatory disorders. The green lines represent all deaths in the United Kingdom, while the blue, yellow, red, and light orange lines represent the number of deaths in England, Scotland, Wales, and Northern Ireland, respectively.

Figure 2The Model 1, representing decision trees, Model 2, representing SVM, and potentially several other models denoted as Model N, undergo training and testing using the provided dataset (heart diseases dataset). Subsequently, their predictions are aggregated using averaging, resulting in a meta-model prediction known as the ensemble prediction.

Figure 3Initially, it partitions the dataset into subsets and assigns equal weight to all data points. Subsequently, a base model is trained on these subsets and utilized to predict outcomes for the entire dataset. The algorithm computes the error by subtracting the predicted results from the actual ones. Once the error is determined, it is given a higher weight for the subsequent observation, aiming to mitigate the error in the next model. This process iterates, with each model endeavoring to rectify the errors of its predecessors. The final prediction is derived by averaging the outputs of the weak models, resulting in the attainment of a robust learner or model.

Figure 4The blue colors show the input layer, the light yellow colors represent the hidden layers while the dark red color reveals the output layer. Every layer in a network has neurons which are visualized by a circle of blue, yellow, and red colors. However, the connection between each neuron and layer is called the weight of the network. The input layer receives the raw information or features that we want the network to process. Each neuron in the input layer represents a feature or attribute of the input data. The hidden layers are responsible for learning and extracting patterns or representations from the input data, while the output layer is where the neural network produces its final predictions or outputs.

Figure 5It utilized the Pearson correlation coefficient to assess the relationship between each feature and the output variable. This statistical test helps us understand if there’s a linear connection between two continuous variables. The coefficient value ranges between –1 and 1, where a value of 1 signifies a strong positive linear relationship, –1 indicates a strong negative linear relationship, and 0 implies no linear relationship between the variables. Therefore, we observed that there are no features highly correlated with each other.

Figure 6Initially, it begins by subjecting the data to preprocessing steps such as investigating the relationships between different features, normalizing the data to ensure consistent scales, and identifying and addressing outliers. Once the data has been preprocessed, we proceed to split it into ten subsets or “folds” using a technique called k-fold cross-validation. In this process, the dataset is divided into ten equally sized portions, and during each iteration, nine of these folds are utilized for training the model, while the remaining fold is used for testing. This process is repeated iteratively, with each fold taking turns as the test set, until all folds have been used for testing. Through this approach, we can evaluate the performance of the model across multiple iterations, which helps to ensure robustness and reliability in our results.

Figure 7The comparasion of machine learning, ensemble learning and deep learning on the basis of accuracy.