A novel approach to the treatment of drug-resistant patients with epilepsy involves the use of implantable devices that deliver electrical stimulation to the epileptic focus at seizure onset. Accordingly, this process requires reliable and energy-efficient seizure detection. To this end, first, for finding the best match between the electrode configuration of an implantable device and the layout of electrodes used during long-term recordings for epilepsy diagnostics, we designed two automatic electrode selection methods. We next implemented four seizure detection algorithms, namely Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and Convolutional Neural Network (CNN). We compared their performance using the automatically selected electrodes. The proposed CNN model showed the best performance, with a mean AUC-ROC (area under the receiver operating characteristic curve) score of 0.94. These results were obtained by applying just four channels with a limited spatial distribution. Therefore, automatic electrode selection methods enable an optimal training of the seizure detection algorithm. Besides, our newly designed seizure detection algorithm is a promising candidate for application in implantable devices.