Acute cardiac rejection remains a significant challenge in the post-transplant period, necessitating meticulous monitoring and timely intervention to prevent graft failure. Thus, the goal of the present study was to identify novel biomarkers involved in acute cardiac rejection, paving the way for personalized diagnostic, preventive, and treatment strategies. A total of 809 differentially expressed genes were identified in the GSE150059 dataset. We intersected genes selected by analysis of variance, recursive feature elimination, least absolute shrinkage and selection operator, and random forest classifier to identify the most relevant genes involved in acute cardiac rejection. Thus, HCP5, KLRD1, GZMB, PLA1A, GNLY, and KLRB1 were used to train eight machine learning models: random forest, logistic regression, decision trees, support vector machines, gradient boosting machines, K-nearest neighbors, XGBoost, and neural networks. Models were trained, tested, and validated on the GSE150059 dataset (MMDx-based diagnosis of rejection). Eight algorithms achieved great performance in predicting acute cardiac rejection. However, all machine learning models demonstrated poor performance in two external validation sets that had rejection diagnosis based on histology: merged GSE2596 and GSE4470 dataset and GSE9377 dataset, thus highlighting differences between these two methods. According to SHAP and LIME, KLRD1 and HCP5 were the most impactful genes.