The ever-growing availability of online curated biological data offers a powerful resource for computational drug repurposing, a strategy that exploits existing drugs for new therapeutic applications. This approach is particularly promising for rare diseases, where the time and cost of traditional drug development pose significant challenges. We present a new drug repurposing methodology that integrates explainable machine learning with open biomedical knowledge graphs. We leverage three such knowledge graphs to construct explainable features based on the concept of shared neighbors between drugs and diseases. These features are then used to train a random forest classifier that predicts potential drug-disease relationships. To validate our approach, we focus on amyotrophic lateral sclerosis (ALS) as a case study. Our methodology demonstrates strong performance on the drug-disease classification task, consistently achieving accuracy above 95% on all datasets. In particular, our explainable feature construction fosters interpretable results, addressing a critical concern in the application of artificial intelligence to biomedical fields. In the ALS use case, our model identified 12 promising drug candidates for repurposing. Significantly, six of these drugs have already been documented in the literature as being associated with ALS, one of them even being the subject of a Phase 2 clinical trial. This work highlights the potential of explainable random forests built on large-scale knowledge graphs for drug repurposing in the context of rare diseases. We show that prioritizing explainability does not compromise prediction quality, even with relatively simple machine learning models. This approach holds great promise for accelerating the discovery of treatments for rare diseases by leveraging the ever-expanding wealth of online biomedical data.