Analysis of the Behavior of The Input Data Set Attributes Affecting the Outputs in MLP Based Artificial Intelligence Models According to the Model

Authors : Muhammer İlkuçar

Pages : 29-37

Doi:10.59940/jismar.1577691

View : 41 | Download : 19

Publication Date : 2025-06-30

Article Type : Research Paper

Abstract :With the widespread use of artificial intelligence, explainability, interpretability, and transparency are very important issues. Especially in the health, defence, security, and law domains. In this study, the same data sets are used for multilayer artificial neural network (MLP) different architectures, and the effects of data set attributes on MLP model output are analysed. The contributions of the model input data attributes to the model prediction were measured with the SHAP method. For the data sets, as the MLP architecture changes, the ranking of the importance levels of the input data set attribute values also changes. It was observed that the change in the attribute influence ranking is mostly valid for attribute values whose contribution levels are relatively close to each other, and the influence ranking of the attributes whose influence ratio is slightly different from other attributes does not change much with the MLP architecture. According to these results, it can be concluded that the model architecture is also effective to a certain extent in Explainable Artificial Intelligence and that there is no relationship between the model\\\'s accuracy value and the importance rates of the attributes.
Keywords : Açıklanabilir Yapay ZekaSorumlu, Sorumlu Yapay Zeka, Derin Yapay Sinir Ağları, Makine Öğrenmesi, SHAP

ORIGINAL ARTICLE URL