Stress is a major health concern, potentially leading to conditions such as cardiovascular disease and anxiety. This underscores the need for effective real-time stress monitoring tools. Photoplethysmography (PPG) sensors, widely available in wearable devices, offer a convenient, cost-effective, and non-invasive method for continuous stress monitoring. This study introduces a convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) based deep learning approach for real-time stress monitoring. The model integrates convolutional layers, max pooling, bidirectional LSTM, LSTM layers, global average pooling, and dense layers. It was trained and validated using the publicly available WESAD dataset, which includes data from 15 healthy subjects. The proposed model achieves an average accuracy of 94.1%, precision of 92.8%, recall of 90.4%, F1 score of 91.3%, and AUC of 0.94. The method demonstrates effective stress detection using only PPG signals with high accuracy, making it a promising tool for real-time stress monitoring and management. This approach highlights the effectiveness of combining CNN, BiLSTM, and LSTM networks in stress classification using PPG signals. It offers significant potential for practical applications in real-time stress monitoring and management.