ارزو صفري

Kidney transplantation is the preferred treatment for individuals with end-stage renal disease. Due to the increasing number of patients in need of transplants, along with the shortage of donor organs and the risk of rejection, one of the most complex aspects of this field is improving the prediction of transplant survival through donor-recipient matching and assigning specific donors to potential candidates on the waiting list. Machine learning can assist doctors and researchers in developing prediction models and identifying crucial factors in graft rejection to predict and enhance transplant outcomes. This study aimed to develop prediction models for short-term kidney transplant outcomes using a comprehensive set of laboratory and demographic data from 657 pairs of recipients and cadaveric donors at Sina Hospital in Tehran between 2014 and 2022. At first, preparation and pre-processing enhance data quality by finding missing data and identifying outliers. To deal with the class imbalance problem, various ensemble learning algorithms, including CatBoost, AdaBoost, RUSBoost, Extreme Gradient Boosting, balanced random forest, and Logistic regression were employed. After fine-tuning the model hyperparameters, crucial features were identified and their impact on the model’s output was determined. Additionally, the optimal threshold for classifying outcomes was established. The voting method was also eva‎luated using three different approaches to improve the model performance. After parameter adjustments, the result showed that the RUSBoost algorithm achieved the highest Area Under the Curve (AUC) with among 79% eva‎luated algorithms. Furthermore, after setting the optimal threshold, the AdaBoost algorithm demonstrated the highest g-means score, exhibiting a 78% improvement compared to the normal state. The selected algorithms with the optimal threshold in the voting model reached a g-means score of 83%.

پيوند كليه , يادگيري ماشين , پيش بيني , انطباق گيرنده و اهدا كننده

Kidney transplant , Machine learning , Prediction , donor-recipient match

Arezoo safari

Dr. Farnaz Barzinpour