ANN-Based Classification of Rain Acoustic Sensor Data Using Modified Mel Frequency Cepstral Coefficients

Tropical cyclones are a common occurrence in the Philippines since it is located in the Western North Pacific region. It is essential to monitor and measure rainfall as it has implications for reducing disaster risk, agricultural planning, and transportation planning. In this study, machine learning artificial neural networks are employed in the classification of acoustic data collected from a rain acoustic sensor (RAS) developed. Mel-frequency cepstrum coefficients are extracted and modified into inputs for an artificial neural network (ANN) that classifies rainfall into Light, Moderate, Heavy, Intense and Torrential according to the Philippine Atmospheric, Geophysical, and Astronomical Services Administration (PAGASA) Rainfall Classification System. Using scaled conjugate gradient backpropagation, the neural network was trained, and its results were compared with the standard tipping bucket rain gauge data. A 94.6% accuracy rate was achieved using five-fold cross-validation for classifying PAGASA’s rainfall data in the experimental study.