In traditional multi-attribute group decision-making (MAGDM) methods, the aggregation of expert evaluation information, the normalization of attribute scales, and the determination of attribute prior weights can lead to information loss and inconsistent results. In addition, these methods cannot identify and improve the shortcomings of non-optimal alternatives. Considering these, this paper proposes a MAGDM method based on data envelopment analysis (DEA) by setting up a parallel expert evaluation system, thus integrating the initial opinions of all experts to rank the alternatives from a global perspective. First, we construct the multi-expert evaluation process as a parallel system from the global perspective so that the decision-making process is based on the initial expert evaluation information. Then, a new directional distance function model based on endogenous projection directions is proposed to measure the alternatives under each expert subsystem, eliminating data processing and artificial weight assignment. The non-optimal alternatives can be improved to an efficient state by adjusting the attribute levels to target values. Next, we propose a DDF super-efficiency model to perform a full ranking of the alternatives. Finally, a case study is provided to demonstrate the rationality and effectiveness of the proposed method.