Active Genetic Learning with Evidential Uncertainty for Identifying Mushroom Toxicity

Abstract

Mushroom's classification as edible or poisonous is an important problem that can have a direct impact on hu-man life. However, most of the existing works do not in-clude model uncertainty in their analysis and suffer from over-confidence issue. To solve this problem, we propose a learning framework, called deep active genetic with evi-dential uncertainty (DAG-EU), to model the uncertainty of the class probability to classify mushrooms. The framework selects the data points with high uncertainty and the most influencing features by using genetic algorithms. The ex-perimental results on the mushrooms dataset demonstrate that the proposed framework can improve the model classi-fication accuracy by 2.3% compared to the methods in the same domain. Moreover, it outperforms the other models from literature by 3.6%.