Electroencephalography(EEG) is the best non-invasive method to record brain activity. Hence, the proposed system MindAlert employs the EEG signal to perform classification for seizure diagnosis. In this MindAlert, we developed a deep neural network namely Metric Learning Based Convolutional Neural Network (MLBCNN). Unlike the conventional CNN, the training process of the MLBCNN is modified according to metric learning which is usually applied in unsupervised learning. The reason for modifying the training process using metric learning is because the EEG waveform of both normal and epileptic seizure in resting state are highly similar to each other. Thus, this cause high data sparsity where the feature embedding vectors of both classes are distributed sparsely and not clustering among the respective class in the metric space. This property causes increases the difficulty in classifying the EEG signal accurately. Hence, as inspired by the metric learning that is usually applied in unsupervised learning, we apply this technique as the training algorithm of the deep neural network CNN as it has the ability to cluster features that are in the same class. The developed MLBCNN outperformed the other machine learning models and the common deep learning models with the best classification performance with an accuracy of 97.4%, AUC of 99.6%, recall rate of 96.9%, precision rate of 91.3%, and specificity of 97.3%. It is the current best deep learning model in seizure EEG signal classification. The MindAlert system can classify the seizure EEG signal and healthy EEG signal within a few seconds. Hence, this invention will hasten the diagnosis process and enlighten the workload of the neurologist. This will definitely benefit not only the doctors but the patients too, as earlier diagnosis may increase the chance to stay in seizure-free life. This brings good news to the nation and globally too as the significant economic impact caused by the epileptic seizure disease can be greatly reduced.