Machine learning for anomaly detection in cyanobacterial fluorescence signals.

Area of Science:

• Environmental science
• Water quality monitoring
• Machine learning applications

Background:

• Harmful algal blooms (HABs) pose risks to drinking water sources.
• Phycocyanin fluorescence monitoring is used, but data interpretation is challenging.
• Existing methods often require cyanobacteria cell counts, which are not always available.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for HAB detection using only phycocyanin fluorescence data.
• To identify effective anomaly detection methods that do not rely on cell counts or biovolume.
• To provide a practical tool for water utilities managing HAB-prone water bodies.

Main Methods:

• Utilized phycocyanin fluorescence data from Lake Erie buoys (2014-2019).
• Applied and compared four open-source machine learning algorithms: Local Outlier Factor (LOF), One-Class Support Vector Machine (SVM), Elliptic Envelope, and Isolation Forest (iForest).
• Trained models on historical data (2014-2018) and tested on independent data (2019).

Main Results:

• One-Class SVM and Elliptic Envelope algorithms demonstrated the highest performance.
• Both models achieved a maximum average F1 score of 0.86 in detecting potential HABs.
• Successful anomaly detection in phycocyanin fluorescence data was achieved without cell count data.

Conclusions:

• One-Class SVM and Elliptic Envelope are promising algorithms for real-time HAB detection.
• These methods offer a viable alternative for water utilities lacking cell count data.
• The study validates the use of machine learning for efficient and accurate HAB monitoring.