drug-disease association analysis via machine learning‎ ‎on extracted‎ ‎features‎ ‎by‎ ‎matrix decomposition

‎drug repurposing presents a cost-effective and time-efficient alternative to traditional drug discovery by identifying new therapeutic uses for existing medications‎. ‎as biomedical data grows in scale and complexity‎, ‎there is an increasing demand for predictive models that balance accuracy‎, ‎interpretability‎, ‎and computational efficiency‎. ‎in this study‎, ‎we systematically evaluate hybrid models that combine established matrix factorization techniques with machine learning regressors‎, ‎with an emphasis on interpretable and lightweight models such as the decision tree regressor‎. ‎using the widely adopted fdataset‎, ‎comprising 1,933 known associations between 593 drugs and 313 diseases‎, ‎we demonstrate that several of these hybrid approaches achieve predictive performance comparable to or surpassing that of complex models like wnmfdda‎, ‎while significantly reducing memory usage and training time‎. ‎notably‎, ‎our framework relies solely on the drug–disease association matrix‎, ‎removing the dependency on auxiliary similarity data‎, ‎which is often unavailable in real-world applications‎. ‎among the tested models‎, ‎the nmf decisiontreeregressor offers the highest accuracy‎, ‎making it ideal for accuracy-critical scenarios‎, ‎while the ridge model stands out for its efficiency and suitability for resource-constrained environments‎. ‎to enhance transparency‎, ‎we further apply lime (local interpretable model-agnostic explanations) to provide interpretable insights into model predictions‎. ‎these findings highlight a practical and scalable framework for drug repurposing‎, ‎particularly suited for environments with limited computational resources‎. ‎our approach supports the development of accessible‎, ‎data-driven predictive tools that accelerate the transition from computational modeling to clinical application‎.