The Concealed Information Test (CIT), a well-established tool in forensic investigations, has thus far been utilised to measure autonomic nervous system (ANS) changes associated with concealed information. While previous studies have explored the integration of eye-tracking technology in face recognition, the specific application of CIT within a mock crime scenario remains relatively uncharted territory. In this study, we aim to broaden the scope of eye-tracking applications using a mock crime scenario, as well as a machine learning classification method to detect hidden crime-related information. Of the four faces displayed as stimuli, the ‘guilty’ group volunteers in the test were able to recognise one as they had previously seen it in the context of the mock crime, whereas the ‘innocent’ group volunteers were all unfamiliar with all four faces. We chose heatmaps depicting the fixation count and fixation durations as the input data for classification. The results obtained with features extracted using ResNet50 and the Support Vector Machine algorithm yielded promising outcomes, achieving an accuracy level of 84.62% for heat maps created using fixation count. These findings suggest the potential development of an innovative tool capable of objectively determining whether an examined person recognises individuals presented in photos, even when denying familiarity with those individuals. The integration of eye-tracking technology and machine learning holds promise for enhancing the accuracy and efficacy of concealed information detection in forensic contexts.