The chaotic categories of hidden big data features, combined with the ignorance of the public key and key encapsulation of big data ciphertext, result in low extraction accuracy and high redundancy of traditional hidden big data feature extraction methods.Here, a secure extraction approach of hidden features of big data is proposed based on hybrid cryptosystem.First, the big data ciphertext is generated through public key encapsulation and cryptographic key encapsulation mechanisms in hybrid cryptosystem.Second, the hidden big data characteristics are categorized based on symmetric encryption and asymmetric encryption designed according to the content of big data ciphertext, which are then used to construct the phase space of big data hidden features and calculate the correlation dimension between big data, thus realize the secure extraction of hidden big data features.The experimental results show that, compared with traditional methods, the proposed approach has low redundancy, high accuracy of classification rate for big data hidden features up to 95%, and low error of feature extraction, verifying the feasibility and application prospect of the proposed approach.