With the increasing penetration of renewable generation integrated into power networks, how to manage uncertainties in optimal power flow (OPF) has become a major concern for network operators. This paper proposes a joint chance-constrained OPF model to tackle uncertainties. This model jointly guarantees the satisfaction probability of all critical OPF constraints so that it can effectively ensure the feasibility of OPF solutions. Considering that the existing works for handling joint chance constraints (JCCs) are either overly conservative or computationally intractable, we propose a time-efficient learningbased robust approximation method for JCCs. It first adopts the sample average approximation (SAA) to convert JCCs into sample-wise constraints with binary variables. Then, the One-Class Support Vector Clustering is introduced to pre-solve the binary variables in SAA. To further improve the computational performance, we design a robust approximation to replace the large number of sample-wise constraints with only a few robust constraints. As a result, the original complex joint chanceconstrained OPF model is formulated into a simple linear form. Moreover, since the proposed model is data-driven, it is applicable to arbitrarily distributed uncertainties. Numerical experiments are conducted to validate the superiority of the proposed method on optimality, feasibility, and computational efficiency.