The density of high frequency data presents a challenge for standard machine learning workflows that lack specialized feature engineering techniques to condense the signal, extracting and highlighting its uniqueness. DataRobot's multimodal input capability supports simultaneously leveraging numerics and images, which for this use-case is particularly beneficial for including descriptive spectrograms that enable us to leverage well-established computer vision techniques for complex data understanding.
This example notebook shows how to generate image features and aggregate numeric features for high frequency data sources. This approach converts audio wav files from the time domain into the frequency domain to create several types of spectrograms. Statistical numeric features computed from the converted signal add additional descriptors to aid classification of the audio source.