Delayed diagnosis of heart failure (HF) is associated with adverse outcomes and increases hospitalisation costs. Early treatment of HF could alter the disease trajectory and reduce clinical events. However, many HF cases remained undetected until advanced symptoms required hospitalisation. Earlier identification and intervention will increase quality of life and mitigate the healthcare burden. Traditional predictive models inadequately captured complex relationships between patient data and HF hospitalisation risk. We aim to develop and validate a novel AI-powered decision support tool, FIND-HF, to predict 1-year HF risk from individuals aged \(\ge \) 40 years using big data from UK primary care health records. The supervised machine learning (ML) algorithms were trained to predict 1-year HF using bootstrap validation. A significant challenge in this study was the rarity of HF events in the dataset, resulting in class imbalance. Among 3,520,186 UK individuals, the incidence of HF and HF hospitalisation was 1.05% at 1 year. For internal validation, logistic algorithm yielded an AUC of 0.786, and LDA and QDA achieved AUCs of 0.787 and 0.782, respectively. The XGB model showed the best discrimination, with an AUC of 0.794. Random forest (RF) exhibited the lowest performance with an AUC of 0.752. The FIND-HF decision support tool successfully identified individuals at risk of incident HF, enabling new approaches for early detection and prevention.