Globally, heart disease continues to claim countless lives, remaining a top killer, highlighting the need for accurate prediction models to aid in early diagnosis and intervention. To enhance heart attack prediction and improve patient outcomes, this study investigates the efficacy of a deep Bi-Directional Recurrent Neural Network architecture. This novel approach leverages stacked Long Short-Term Memory (LSTM) units to capture temporal dependencies within medical data, potentially surpassing the accuracy of conventional Machine Learning (ML) algorithms in predicting heart attacks. The results highlight that the twin Long Short-Term Memory model achieved an exceptional accuracy of 98% in predicting heart attacks, surpassing other conventional ML algorithms. The results highlight the significance of utilizing sophisticated machine learning techniques in the medical field to enhance patient results and decrease the prevalence of cardiovascular illnesses.