Epileptic seizure is a chronic neurological disorder which affects millions of people all over the globe. It can be treated in a better way if the symptoms are detected at an early stage. In this study, we have demonstrated and evaluated the classification performances of different machine learning classifiers for the detection of epileptic seizures from electroencephalography (EEG) signals. For this, we have first applied principal component analysis (PCA) on EEG signals to obtain much reduced-length PCA vectors. These vectors are then applied to decision tree (DT), k-nearest neighbor (KNN), Naïve Bayes (NB), support vector machine (SVM) and artificial neural network (ANN) classifiers for the detection of epileptic seizures. The effects of length of PCA vectors on the performances of these classifiers have also been analyzed rigorously for 2-class, 3-class and 5-class classification of EEG signals. Besides such PCA-based classifiers, we have also proposed and evaluated the performances of a customized convolutional neural network (CNN) to directly extract features from the EEG signals as well as to perform classification tasks. The results showed that CNN outperforms PCA-based machine learning classifiers. For 2-class classification cases, CNN attains classification accuracies in the range from 99.50% to 100%, whereas 98.48% and 96.32% accuracies are obtained with CNN for 3-class and 5-class classification cases. The results signify that the proposed CNN classifier can be considered as a highly-efficient scheme for the reliable detection of epileptic seizures from EEG signals. Bangladesh Journal of Medical Physics Vol.15 No.1 2022 P 28-42