Purpose <p>Cancer is the second leading cause of death globally, with a rising incidence each year. While imaging-based techniques like CT and MRI are considered diagnostic gold standards, they often detect cancer only after tumour development. This delay is particularly concerning in countries where access to advanced medical infrastructure is limited. Such healthcare disparities hinder timely detection and ineffective treatment, which we are trying to solve with this study.</p> Methods <p>We have targeted epigenetic variations in cell-free DNA (cfDNA) found in blood plasma, observed at least 2–3 years before the tumour develops and symptoms occur. In this paper, a model is devised using clinical data from hospitals located in New Delhi, India, to detect cancer and its stage. The dataset consists of 109 cancer patients with 10 different cancer types and 50 healthy individuals. The analysis utilized ten classification models to estimate the patient’s prognosis. Further, feature selection of the best-performing model was devised by SHAP analysis.</p> Results <p>The experimental results demonstrated that the Light Gradient Boosting Machine (LightGBM) classifier is superior to other models on a performance basis that providing the best results over other classifiers with precision, recall, and F₁ scores of 0.91, 0.95, and 0.91, respectively. The top features- cfDNA concentration, impedance signals, and global methylation percentage were found for the LightGBM classifier.</p> Conclusion <p>Finally, a platform created can accurately predict the presence and stage of cancer within a few hours. It is non-invasive, affordable, and also has applications in treatment assessment and regulation.</p>