Benchmarking Machine Learning Algorithms for Epilepsy Detection on Multi-Age Datasets

Epilepsy is a neurological disorder characterized by abnormal electrical activity in the brain that can occur in individuals of any age. Electroencephalography (EEG) is a common method for diagnosing epilepsy. In EEG analysis, doctors examine different characteristics of the EEG signal, such as waveform, frequency, amplitude, and intervals, to determine whether abnormal electrical activity is present. This study aimed to detect epilepsy in patients of all ages based on EEG signals using various machine learning methods. This study included four publicly available datasets covering infants, children, and adults. A total of 23 features were extracted, including time domain, frequency domain, and age features. We used 4 data sets to train and test five machine learning models and evaluate the performance of the five machine learning models. Experimental results show that multi-layer perceptron (MLP) achieves a good balance between recall and accuracy in identifying epilepsy in infants, children, and adults. They show good generalization ability in distinguishing epileptic and non-epileptic EEG signals in epilepsy datasets of different age groups. The area under the curve (AUC) values exceed 0.9, which further confirms their effectiveness. Furthermore, by performing feature importance analysis using the random forest method, it was found that age-related features play an important role in epilepsy detection.