Thyroid disease diagnosis based on feature interpolation and dynamic weighting ensemble model

Authors:

Abstract
Medical diagnosis is the basis for modern medical treatment planning, rapid and accurate testing is responsible for patients’ lives and health. Thanks to increasingly sophisticated large-scale data collection techniques, the advantages of big data and artificial intelligence technologies for data analysis are gradually coming to the fore. However, medical data mostly have missing items and non-human biases, and common algorithms often show instability in learning data distribution patterns. In this paper, a feature interpolation algorithm based on feature importance is proposed for thyroid disease classification dataset, and a dynamic assignment integration model is designed to further optimize the weight assignment of the integration model in terms of distance and accuracy, which improves the accuracy and smoothness of the model. The classification accuracy reaches 94.75%, which is 2.45% higher than the original model, and the index AUC is almost the same as the original model. The generalization learning performance is poor because the deep learning algorithm lacks the support of a priori knowledge in learning data features. In this paper, the convolutional model is introduced to the sequence data prediction and propose a feature-enhanced small network EdcNet, and combine it with the feature interpolation algorithm to explore the role and effect of lightweight convolution in sequence analysis, and the final classification accuracy and AUC reach 95.6% and 99.1%, respectively.

This publication has 14 references indexed in Scilit: