Background <p>This study focuses on detecting mental performance from EEG signals. It provides both classification and explanation results. For this purpose, we developed a new feature extraction method called Different Pattern (DiffPat) within an Explainable Feature Engineering (XFE) framework.</p> New method <p>The proposed approach utilizes an EEG mental performance dataset. The DiffPat algorithm extracts the differences between EEG channels in order to improve the efficiency of the feature extraction process. To develop the identified features into a more accurate set of features, the iterative Neighborhood Component Analysis (INCA) feature selection method was utilized iteratively. After selecting the features with INCA, the K-Nearest Neighbor (KNN) classifier was used to classify the features. AI-based explainable results utilizing XAI methods in conjunction with Directed Lobish (DLob) symbolic language were generated to produce cortical connectome diagrams and interpretive sentences that describe the relationship between the brain’s neural networks.</p> Results <p>The DiffPat-driven model achieved 84.48% accuracy with leave-one-subject-out (LOSO) cross-validation, which serves as the primary subject-independent performance estimate. A complementary subject-aware 10-fold cross-validation, where all segments of each subject were confined to a single fold, yielded 99.87% accuracy.</p> Comparison with existing methods <p>The DiffPat-based XFE model can explain the results and provides an advance in both feature engineering and neuroscience.</p> Conclusions <p>Because the model can both classify and explain the results, it has strong potential for mental performance analysis and other EEG applications.</p>