Despite advances in managing the condition, heart disease remains a significant public health burden worldwide. Given this context, early prediction and intervention are essential. Given the complicated medical data available, practitioners now approach heart disease prediction differently with the help of advanced selections for ML (Machine Learning). This paper provides a survey on different ML techniques developed for the diagnosis of heart disease, covering new methodologies, performance evaluation metrics and challenges which are used in heart disease prediction. This paper uses more than thirty journal papers to survey different ML models from the traditional algorithms to deep learning (DL) approaches. They also discuss improvements on prediction accuracy, explainability challenges, and the use of multimodal data. The paper closes with a series of future research recommendations and an exploration of possible betterment in the ML-based prediction models on heart disease.