Interictal Epileptiform Discharges (IEDs) are essential biomarkers of epilepsy in patients’ electroencephalograms (EEGs).The correct detection of IEDs helps the diagnosis, the localization of the epileptogenic zone and the monitoring of treatment. The gold standard to detect IEDs is the visual inspection of EEGs, performed by clinical experts. However, this task is time-consuming, prone to human error and subjectivity, and suffers from intra and inter-observer variability.This motivates the development of automated IED detectors. Researchers have explored various methods, from classical signal processing to end-to-end machine learning (ML) models. In recent years ML algorithms have gained in popularity for their high accuracy, often outperforming traditional signal processing techniques and achieving performance levels comparable to human experts.Yet, critical issues remain. ML algorithms are typically trained on private databases, making direct comparison across studies difficult, and publicly available datasets are often limited in size and diversity. EEG patterns vary significantly across individuals and even within the same patient, making it difficult for ML models struggle to generalize across multiple EEG profiles. There is also a high variability in the recording conditions (duration, number of channels, sampling rate) making the generalization of ML models complicated.To overcome these hurdles, we propose individualized ML models tailored to each patient’s unique EEG characteristics. Using a limited amount of annotated data, we show the importance of the preprocessing steps (dataset preparation). By testing different model architectures, our preliminary results demonstrate that well-configured individualized detectors provide valuable insights into IED quantification and localization from EEG signals.