P300-based deception detection of mock network fraud with modified genetic algorithm and combined classification

Abstract

To detect network fraud, a three-stimulus paradigm was used in a mock crime P300-based concealed information test. A P300-based deception detection method based on a modified genetic algorithm and a confidence-coefficient-based combined classifier was created for mock network fraud detection. After the multi-domain integrated signal preprocessing and feature extraction, a modified logistic equation based multi-population genetic algorithm was adopted for feature selection to obtain an optimal feature subset. Then the confidence coefficient was proposed to determine the classification difficulty levels of samples. A combined classifier based on confidence coefficient was proposed for classification. Compared with the component classifiers and other individual classifiers, the combined classifier requires 34% less computing time and the mean classification accuracy rate is 0.2 to 2.23 percentage points higher for twelve subjects using leave-one-out cross validation. Experiment results confirm that the proposed method is effective to detect deception during network fraud simulation.