In recent decades, the use of email has become widespread, leading to SPAM or fraudulent messages. Artificial intelligence (AI) and especially machine learning (ML) is a promising solution to classify these messages into two categories: SPAM and HAM (legitimate). However, this classification approach shows unsatisfactory performance due to the low success rate of valid message classification. To improve this situation, we propose in this article optimal architectures to efficiently detect SPAM and HAM. Optimizers are applied to CNNs (Convolutional Neural Networks), ANNs (Artificial Neural Networks), and LSTM (Long Short-Term Memory) basic architectures as well as other ML algorithms such as SVM (Support Vector Machine), DT (Decision Tree), and NB (Naive Bayes). The experimental results showed that applying optimizers with basic CNNs architecture outperforms several classifiers by achieving an accuracy of 0.99%. The proposed optimal architecture has the potential to efficiently detect SPAM from HAM.