Optimized CNN-RNN architecture for rapid and accurate identification of hazardous bacteria in water samples

• Innovative Approach: Proposes an optimized CNN-RNN architecture that combines spatial feature extraction (CNN) with temporal growth pattern analysis (RNN) for rapid bacterial detection in water samples. • High Accuracy: Achieves 97.51% accuracy with stained bacterial images and 96.23% accuracy with unstained images, demonstrating its efficacy across various conditions. • Improved Sensitivity and Specificity: The model offers 98.57% sensitivity and 94.89% specificity for stained images, ensuring reliable bacterial identification and minimizing false negatives and positives. • Real-Time Monitoring Potential: Opens possibilities for real-time water quality monitoring when integrated with IoT systems, addressing critical water safety issues. • Comprehensive Evaluation: Includes a performance comparison with conventional methods (CNN, RNN, ANFIS), showing superior results in bacterial classification and identification. Drinking water safety is a critical global issue, as pathogenic bacteria in water can cause various severe diseases, including diarrhea and systemic infections. Rapid and accurate detection of hazardous bacteria is key to ensuring water quality, especially in regions with limited access to water treatment facilities. Conventional detection methods, such as bacterial culture, are often time-consuming and may not detect bacteria in the "viable but non-culturable" (VBNC) state. To address these limitations, this study proposes the development of an optimized Convolutional Neural Network-Recurrent Neural Network (CNN-RNN) model for identifying harmful bacteria in drinking water samples. The CNN is used to extract spatial features from microscopic bacterial images, while the RNN handles temporal patterns in bacterial growth, enabling the system to detect bacteria more accurately. Experimental results show that the model, when using bacterial image staining, achieved 97.51% accuracy, 98.57% sensitivity, and 94.89% specificity. Even without image staining, the model still performed well, with 96.23% accuracy and 98.89% specificity. These findings indicate that the optimized CNN-RNN model can provide an efficient and rapid solution for detecting hazardous bacteria in drinking water. This research paves the way for further development, including the integration of IoT for real-time water quality monitoring.