Predicting kidney transplantation outcome based on hybrid feature selection and KNN classifier

Abstract

Kidney transplantation outcome prediction is very significant and doesn’t require emphasis. This will grant the selection of the best available kidney donor and the best immunosuppressive treatment for patients. Survival prediction before treatment could simplify patient’s decision making and boost survival by altering clinical practice. This paper proposes a new novel prediction method based on data mining techniques to predict five-year graft survival after transplantation. This new proposed prediction method composes of three stages: data preparation stage (DPS), feature selection stage (FSS), and prediction stage (PS). The new proposed prediction method merges information gain with naïve Bayes and k-nearest neighbor. Initially, it uses information gain to select the essential features, uses naïve Bayes to select the most essential features. These two methods are combined in a new hybrid feature selection method which chooses the minimum number of features that produce highest accuracy. Finally, it uses k-nearest neighbor for graft survival prediction classification. The proposed prediction method has been evaluated against recent techniques. Experimental results have proven that the proposed prediction method outperforms the recent techniques as it attains the maximum accuracy and F-measure with minimal errors. This prediction method can also be used in other transplant datasets.