Abstract Background Heart failure (HF) is an important long-term complication in cancer survivors, especially in breast and haematological malignancies treated with anthracylines. However, screening for left ventricular dysfunction in all survivors is impractical due to potential low cost-efficiency. So we sought to develop a cancer-specific risk scoring system for HF (CRS-HF). Methods Cancer survivors (n=43,720) participating in the UK-Biobank were evaluated in this study (age 59.9 ± 7.4 years, 62.4% female). Participants were randomized into training and testing dataset with a 7:3 split. A Random Forest algorithm was applied to identify key variables that are predictive for HF incidence among cancer survivors. Results Incident HF developed in 4.3% over a median of 12.9 years. The random forest algorithm identified demographic (age, sex and body mass index), clinical (heart rate, blood pressure and cardiac history), pathology (lipids) results and other patient reported outcome measures (PROMS) that predicted HF with an accuracy of 0.872 (95%CI: 0.867-0.877) in the training set. In the testing set, the accuracy was 0.840 (95%CI: 0.832-0.848) and a reported sensitivity and specificity of 62.42% and 84.93% respectively. Our study also demonstrated that the CRS-HF performed better in ruling out non-HF incidence among survivors from breast cancer and lymphoma (accuracy: 0.858) than other cancer types (accuracy: 0.836). Conclusion The CRS-HF score is a prove-of-concept that appears to be an effective means of identifying HF risk in cancer survivors. Although initial findings demonstrate promising accuracy in forecasting HF occurrences among cancer survivors, comprehensive validation and potential refinements are warranted across diverse population cohorts.