Improving the technical efficiency of agricultural production is an important part to promote the high-quality development of agriculture. However, in practical application, there exist some flaws in the traditional technical efficiency evaluation model based on the frontier, such as slow computing speed and low flexibility, which make it difficult to evaluate the efficiency of a large number of new samples. For the above reasons, a method for evaluating and predicting the technical efficiency of agricultural production was proposed, which combined the DEA technical efficiency measurement model based on the frontier with the ensemble learning model, and the grape production technical efficiency dataset was used to verify the effect of the model. Experiments showed that the Stacking fusion model reached the accuracy and AUC of 94.8% and 0.984 respectively, with promising result that surpassed the other comparison models, indicating that the Stacking ensemble learning model had high accuracy, robustness and generalization ability, and can achieve more efficient, fast and stable technical efficiency evaluation.