Voice disorder is an early symptom of Parkinson’s disease (PD), which makes acoustic analysis be used to evaluate voice performance from the early stage of PD. However, the existence of confounding factors in PD dataset leads to contradictory conclusions in different studies. In order to reduce the influence of confounding factors and improve the stability and generalization ability of acoustic measurement feature, congruity features across datasets (CFAD) based on formal concept analysis (FCA) are proposed in this study. Firstly, the significant features in different datasets are selected by using CART; Then, the formal context between the datasets and the significant features is established based on FCA; Finally, the formal concepts are extracted based on the formal context to obtain the combination of significant features under the different combination of popular datasets, completing the extraction of CFAD and verifying the performance. The results have shown that CFAD have achieved more than 90% of the performance of the original dataset on four datasets. It can be concluded that CFAD provide a reference value for the study of acoustic measurement features in the field of PD, helping to establish a more reliable PD detection model with higher generalization.