An evaluation method for the prediction effect of classification models based on deep learning
Tao Zhang, Yifei Xu, Ling Tan, Faling Li, Miao Song, Shiyu Zhu · 2023
In the field of machine learning, we realize that success does not depend solely on the structure and number of parameters of the model, but also on model evaluation. In cla-ssification tasks, when faced with unlabeled data, traditional methods can no longer accurately assess the confidence level of model predictions. Therefore, this paper proposes a novel evaluation method to evaluate the difference between the model prediction results and the clustering results by selecting key features and clustering to measure the confidence of the model in predicting unlabeled data. With this approach, we were able to accurately assess the model's confidence in the predictions of unlabeled data and experimentally demonstrated its feasibility. At the same time, this paper also expands the field of research beyond binary classification cases, but covers more general m-ulticlassification cases.