Advancement of technology has caused the production of more data, both in terms of volume and dimensions, which this increase affects the machine learning model Dash and et al. (2019). Therefore, it is necessary to implement new algorithms to understand this volume of data. One of the popular mutual information-based approaches is the Minimal-RedundancyMaximal-Relevance criterion (MRMR) Peng and et al. (2005), which considers feature relevance concerning class labels and ensures that redundant features are not present in the final feature subset. The MRMR algorithm evaluate bivariate feature dependencies. Moreover, the current algorithm is susceptible to transformations, such as scaling. To address this problem Lall and et al. (2021) proposed a feature selection algorithm based on copula, which on several real and synthetic datasets, the proposed algorithm performed competitively in maximizing ∗Corresponding author, emadi@um.ac.ir 1 Stability Performance of CBFS Algorithm 2 classification accuracy. in this article, we present Copula Based Feature Selection (CBFS), a filter based forward sequential search technique for feature selection, which In a real data set, the proposed algorithm performed in maximizing classification accuracy. CBFS works by minimizing the copula mutual information among selected features and maximizing the same between a candidate feature and class, And that this method does not depend on the data set. We have shown that the CBFS algorithm performs more stably than the MRMR algorithm in selecting features from noisy data. Section 2 presents the main concepts. Section 3 presents Copula Based Feature Selection Algorithm and minimal-redundancy-maximal-relevance Algorithm and Stability Performance. In Section 4, we have simulated the stability of the CBFS method in feature selection from noisy data on a real data set. 2 Concepts In this section, we show some general concepts about copulas and mutual informatio

کلمات کلیدی