Kidney transplantation is the best treatment available for most patients in the chronic renal failure stage, which has been associated with reduced mortality, improved quality of life, and cost compared to chronic dialysis in most patients. However, there is always a risk of mismatch between the recipient and the kidney donor. Prediction of preoperative adaptation can help to increase the survival rate of renal transplant patients. Clinical decisions before kidney transplantation to predict compliance, although it is done with many tests, are based on doctorsʹ reasoning and this process can lead to the wrong decision. Using data mining and machine learning methods is an effective step towards reducing these mistakes and can improve the decision-making process. In this study, an internal dataset related to kidney transplantation was used that has information about donors and recipients from before transplantation up to 6 months after transplantation. According to the performance indicators related to the eva‎luation of kidney function status and the resulting indicators including rejection or acceptance of transplantation in the data set used, two regression problems (for functional indicators) and classification (for the result indicators) are studied. In the regression section, the effective parameters in the success of kidney transplants are categorized and ranked to improve kidney transplantation function. In the classification section, the kidney transplant result are predicted using different methods of machine learning and according to the information available from the dataset. The results of this study show that clinical and non-clinical parameters are of particular importance in predicting the outcome of kidney transplantation. Another result is that different methods of dimension reduction and data imbalance control methods, which are among the methods of data preprocessing, can improve the process of predicting kidney transplant outcomes. In the regression section, the support vector machine algorithm with linear kernel function obtained the best-estimated result with the criterion MAE equal to 8.60. In the classification section, the random forest algorithm, along with the feature extraction method LDA on balanced data using oversampling, obtained the accuracy of 0.9776, precision of 0.97, and recall score of 0.97.

مسعودرضا هاشمي، زينب زالي

مهران صفاياني