Spam email has become one of the most annoying topics in our daily lives. On the other hand, an unsolicited bulk email is not legitimate when accessing a user account without permission. For this reason, detecting spam is important to protect against unwanted messages and emails. Several studies have been carried out to solve the problem using different techniques. In the study, we examine the effectiveness of the dropout strategies of recurrent neural networks (RNN) algorithms such as long short-term memory (LSTM), bidirectional LSTM (BiLSTM), and gated recurrent unit (GRU) to minimize the overfitting problem and track down the best performance compared to other models. In terms of data preprocessing, we used Synthetic Minority Oversampling Technique (SMOTE) techniques to analyze the balanced data. As a result, compared to the strategies, recurrent dropout and spatial dropout strategies achieved the highest testing accuracy of 97.48% and training accuracy of 99.18%. The outcome shows the confusion matrix and ROC curve that perform the best value, which is nearest to value 1. The study could be used to improve the preprocessing techniques and effectiveness of dropout strategies in recurrent neural networks for other text classification.