A Machine learning based data analysis was performed for the data analysis obtained from a low-cost alternative, smartphone based, portable electrochemiluminescence (ECL) sensor for phenolic compounds. While the phenolic compounds, vanillic acid and p-coumaric acid, effectively quench the ECL reaction, the light intensity and electric current were simultaneously recorded by the smartphone apps. Due to common problems present in sensor data such as non-linearity, multimodality, sensor-to-senor variations, presence of anomalies, and ambiguity in key features, several machine learning strategies were explored. In contrast to the traditional calibration approach of extracting predetermined key features, the ML methods such as tri-layer neural net or boosted trees carried out effective regression tasks by learning higher patterns without processing the key features. Combined multimodal characteristics made 80% enhanced performance with multilayer neural net algorithms than the traditional approaches. The results demonstrated that the ML methods could provide robust analysis framework for sensor data with noises and variability without preprocessing to extract features or examine ambiguous anomaly.
Available at: http://works.bepress.com/hyun_kwon/37/
National Science Foundation CBET division (#1706597)