Hadith is the secondary source of Islam legislation that has three-part, i.e., Sanad, Matn, and Taraf. Sanad is an essential part of Hadith that represent the chain of Narrator who conveys the Hadith. Based on Hadith Science, the authentication of the Hadith also could be observed through the state of its Sanad. Most studies on the Hadith Sanad representation apply an ontology and XML. Thus, this study proposed a new model of Sanad Hadith representation exploits the Graph model. First, the candidate of Graph node and Graph relation were extracted automatically from raw Arabic Hadith text using Arabic Part of Speech (A-POS) and Arabic Named Entity Recognition (A-NER). Then, a novel machine learning model for the Hadith Sanad Graph Construction developed employs SVM and GBM algorithm. That model attained the best performance on 0.84 and 0.92 precision average, 0.83 and 0.91 recall average, 0.82 and 0.90 f1-score average. The final result of this study was a Hadith Sanad Graph that had been verified the correctness compare with the original Hadith text.