Anomaly detection in earth dam and levee passive seismic data using support vector machines and automatic feature selection

We investigate techniques for earth dam and levee health monitoring and automatic detection of anomalous events in passive seismic data. We have developed a novel data-driven workflow specific to our domain, which could be generalized for monitoring other systems with time series data. We use machine learning and geophysical data collected from sensors located on the surface of the levee to identify internal erosion events. In this paper, we describe our research experiments with two-class and one-class support vector machines (SVMs). We use two different data sets from experimental laboratory earth embankments (each having approximately 80% normal and 20% anomalies) to ensure our workflow is robust enough to work with multiple data sets and different types of anomalous events (e.g., cracks and piping). We apply wavelet-denoising techniques and extract nine spectral features from decomposed segments of the time series data. The two-class SVM with 10-fold cross validation achieved over 94% overall accuracy and 96% F1-score. Experiments with the one-class SVM (no labeled data for anomalies) using the top features selected by our automatic feature selection algorithm increase our overall results from 83% accuracy and 89% F1-score to over 91% accuracy and 95% F1-score. Results show that we can successfully separate normal from anomalous data observations.