The invention relates to the technical field of
deep learning, in particular to an input
malaria related marker mining method and device, and the method comprises the steps: constructing a sample through a plurality of input
malaria features, carrying out the multiple training of an input
malaria prediction model, and carrying out the explanation of the trained model through a plurality of interpreters; aggregating the local interpretation results obtained by each
interpreter in different training batches into a global interpretation result, and obtaining the aggregation weight of each
interpreter according to the reproducibility and feature authenticity of the
interpreter; and obtaining an attribution
score and an importance sequence of each input malaria feature according to the global interpretation result and the aggregation weight of each interpreter. According to the method, multiple interpreters are integrated based on consistency and information
perception to obtain the interpretation result, the consistency, accuracy and robustness of the interpretation result are improved, researchers are helped to find
functional features having important significance on input malaria, and the accuracy of input malaria prediction is improved.