Article Info
Heart Disease Prediction Using Artificial Neural Network with ADAM Optimization and Harmony Search Algorithm
Alyaa Ghazi Mohammed, Mohd Zakree Ahmad Nazri
Abstract
Heart diseases represent a leading global health concern, underscoring the imperative for innovative strategies in early detection and prevention to effectively mitigate risks and avert sudden fatalities. The intricate nature of cardiac function demands a robust analytical framework capable of processing vast, multidimensional datasets while prioritizing critical features that significantly influence the prediction of heart health outcomes. This study introduces a multi-layer perceptron neural network (MLP) algorithm tailored to predict the likelihood of coronary artery disease (CAD) onset by meticulously analyzing relevant risk factors derived from the Z-Alizadeh Sani dataset, a comprehensive repository of clinical data that captures diverse patient profiles and diagnostic indicators. Drawing from an extensive review of existing predictive models and cardiovascular health risk factors, this research proposes an enhanced ADAM optimization algorithm, integrated with advanced data processing and feature selection methodologies, to identify and refine key predictors for improved model performance. The ADAM optimizer effectively tackles challenges in continuous parameter optimization by dynamically updating the model's weights and biases, adapting the learning rate for each parameter based on accumulated historical gradient information to achieve more efficient minimization of the loss function during training. Complementing this, the Harmony Search Algorithm (HSA) is incorporated to augment data features, facilitating better pattern recognition and enhancing overall classification accuracy through optimized feature engineering. Our in-depth analysis underscores the substantial relevance of the Z-Alizadeh Sani dataset in accurately categorizing heart disease manifestations, with the proposed CAD model achieving a competitive accuracy rate of 86.66% when evaluated on subsets from the UCI repository. This performance is validated through rigorous comparative assessments against various classification algorithms and state-of-the-art methods, revealing notable advantages in terms of predictive precision, computational efficiency, and adaptability to real-world clinical scenarios. In summary, this study advances the field by delivering an effective, optimized predictive algorithm for early heart disease detection, thereby offering valuable insights that could enhance healthcare outcomes, support proactive cardiovascular risk management, and pave the way for future innovations in personalized medicine
keyword
Feature selection, Biomedical data, Discrete Binary Harmony search, Optimization of ANN, ADAM optimization
Area
Pattern Recognition