In today’s world, rapid developments in science and computers are increasingly adding up to larger amounts of data; as a result, numerous problems have emerged in the analysis of big data. Hence, data dimensionality reduction can accelerate data analysis and even yield better results without losing any useful data. This study proposes a novel copula-based method of dimensionality reduction and then compares it with the principal component analysis (PCA), which is usually adopted in data mining. And we show the superiority of the new method. A copula represents an appropriate model of dependence to compare multivariate distributions and better detect the relationships of data. Therefore, a copula was employed in the proposed method to identify and delete noisy data and data of many common features from the original data. Applied to the reduced data, classification methods are finally compared.