Related Experiment Video
Updated: Jul 26, 2025
Enhanced Reproducibility and Precision of High-Throughput Quantification of Bacterial Growth Data Using a Microplate Reader
Published on: July 27, 2022
Predicting heterotrophic plate count exceedance in tap water: A binary classification model supervised by
Ji Won Park1, Joby Boxall2, Sung Kyu Maeng1
1Department of Civil and Environmental Engineering, Sejong University, 209 Neungdong-ro, Gwangjin-gu, Seoul 05006, Republic of Korea.
This study explores how machine learning can predict when drinking water may have high levels of bacteria, using data that doesn't rely on traditional lab cultures. The researchers used a model called an artificial neural network to analyze data from water samples. They found that measures like intact cell count and ATP could reliably predict when bacterial levels would exceed safe limits. The model worked well, with high accuracy and minimal false alarms. This approach could help improve water safety by providing faster, more reliable results than current methods.
Area of Science:
- Environmental microbiology
- Water quality monitoring
- Machine learning in public health
Background:
Drinking water safety relies on heterotrophic plate count (HPC) as a standard indicator of microbial quality. However, HPC represents a small fraction of the total bacterial community and has delayed response times. Culture-independent methods like intact cell count (ICC) and adenosine triphosphate (ATP) offer more immediate insights. Prior research has shown correlations between these measures and HPC, but no predictive model has yet been developed. This gap motivated the use of machine learning to explore nonlinear relationships. No prior work had resolved how to translate ICC and ATP data into HPC predictions. The study aimed to address this limitation by applying artificial neural networks. Existing knowledge suggested that ICC and ATP correlate with HPC, but the exact nature of these relationships remained unclear. The study sought to clarify how these variables could be used to predict HPC exceedances in tap water.
Purpose Of The Study:
The aim of this study was to develop a predictive model for heterotrophic plate count (HPC) exceedance using culture-independent data. The researchers wanted to explore whether ICC, ATP, and chlorine concentrations could reliably predict HPC levels in tap water. HPC is a standard indicator of water quality but has limitations due to its small proportion in the microbial community and delayed response. The study aimed to overcome these limitations by using machine learning. The motivation came from the need for faster, more reliable water quality assessments. By using ICC and ATP, which provide immediate data, the model could offer near real-time predictions. The study also aimed to validate the nonlinear relationships between HPC and ICC/ATP. The goal was to create a binary classification model that could detect HPC exceedances with high accuracy. This would help improve the biostability and safety of drinking water.
Main Methods:
The study used a 2-layer feed-forward artificial neural network (ANN) to classify HPC exceedance in tap water samples. Input variables included intact cell count (ICC), adenosine triphosphate (ATP), and free chlorine concentrations. The model was trained on stagnant and flushed water samples to capture variability. Nonlinear relationships between HPC and ICC/ATP were confirmed through statistical analysis. The dataset included both stagnant and flushed conditions to test model robustness. The ANN was optimized for binary classification of HPC exceedance. Feature importance was assessed to determine which variables most strongly influenced predictions. The model's performance was evaluated using accuracy, sensitivity, and specificity metrics.
Main Results:
The best binary classification model achieved an accuracy of 95%, sensitivity of 91%, and specificity of 96%. Intact cell count (ICC) and free chlorine were the most important features for classification. The model successfully predicted HPC exceedance in both stagnant and flushed water samples. Nonlinear relationships between HPC and ICC/ATP were confirmed through the model's performance. The study demonstrated that ICC and ATP data could reliably predict HPC levels. The model's high specificity suggests it minimizes false alarms in water quality monitoring. The results indicate that culture-independent data can be used to replace traditional HPC testing. This approach offers near real-time data for improved water safety assessments.
Conclusions:
The study demonstrated that a machine learning model can predict HPC exceedance using culture-independent data. ICC and ATP provided reliable inputs for the binary classification model. The model's high accuracy and specificity suggest it could replace traditional HPC testing. This approach offers faster, more reliable water quality assessments. The results support the idea that ICC and ATP can be used to monitor biostability in drinking water. The model's performance indicates that nonlinear relationships between HPC and ICC/ATP are significant. The study also highlighted limitations such as sample size and class imbalance. The findings suggest that culture-independent data can improve the safety and biostability of drinking water.
Frequently Asked Questions
The model achieves 95% accuracy in predicting HPC exceedance using ICC, ATP, and chlorine data.
ICC was found to be the most important feature for predicting HPC exceedance in the study.
The model uses a 2-layer artificial neural network to capture nonlinear patterns in the data.
Free chlorine concentrations were among the most important features for predicting HPC exceedance.
The model has a sensitivity of 91% in detecting HPC exceedance.
The study notes limitations such as sample size and class imbalance affecting model performance.
