RoLA++: A Lightweight Adaptive Framework for Real-Time Anomaly Detection in Marine Sensor Data This record contains the draft manuscript, supplementary reports, and supporting materials for a study on lightweight real-time multivariate anomaly detection in marine sensor streams. The paper extends the RoLA (Real-time Online Lightweight Anomaly Detection) framework with practical enhancements for improved robustness and operational use in sensor-rich environments. The study focuses on streaming multivariate marine telemetry (e.g., temperature, turbidity, flow, optode signals, conductivity, and salinity) and evaluates RoLA++ under realistic online conditions. In addition to detection quality, the evaluation emphasizes deployment-relevant metrics such as latency, throughput, memory footprint, false alarms, and robustness to missingness and drift. Key contributions include: - A lightweight adaptive extension of RoLA for real-time multivariate anomaly detection - A streaming evaluation protocol for marine telemetry with operational performance metrics - Comparative benchmarking against baseline detectors (e.g., robust z-score/MAD, EWMA, CUSUM, online PCA residual, and batch baselines) - Robustness analysis under injected missingness and drift scenarios - Reproducible experimental assets (reports, code links, and configuration details where applicable) This upload is intended for research collaboration, review, and versioned archiving. The manuscript may be updated as figures, additional experiments, and final revisions are completed. Keywords: anomaly detection, streaming analytics, marine sensor data, multivariate time series, lightweight AI, online learning, environmental informatics, RoLA
Orokpo et al. (Mon,) studied this question.