This thesis presents a fully automated, multi-stage pipeline to localize face-sensitive cortical sites from stereo-EEG (SEEG) recordings. The workflow begins with robust artifact removal to improve the signal-to-noise ratio, followed by feature extraction that captures complementary aspects of the neural response—event-related potentials (ERPs), time–frequency representations via wavelet transforms (WAV), and image-based encodings such as Gramian Angular Fields (GAF). Finally, we train machine-learning models on these features to classify responses to faces versus other stimuli and to generate spatial maps that highlight electrode contacts likely involved in face processing. Taken together, the pipeline provides an end-to-end, data-driven approach for identifying functionally relevant cortical areas in clinical SEEG, with a focus on regions supporting social visual perception.