2025_programme: Assessment of Decision Tree and Multilayer Perceptron Performance in Seabed Characterization

• Day: June 17, Tuesday
    Location / Time: B. ERATO at 11:40-12:00
• Last minutes changes: -
• Session: 19. Uncertainty quantification and machine learning in signal processing
  Organiser(s): Angeliki Xenaki, Zoi-Heleni Michalopoulou
  Chairperson(s): Angeliki Xenaki, Zoi-Heleni Michalopoulou
• Lecture: Assessment of Decision Tree and Multilayer Perceptron Performance in Seabed Characterization [Invited]
  Paper ID: 2153
  Author(s): Diego Rios, Zoi-Heleni Michalopoulou
  Presenter: Zoi-Heleni Michalopoulou
  Abstract: Understanding the oceanic propagation medium, particularly seabed properties, is critical for accurate source localization, especially when detecting weak targets. This study utilizes machine learning techniques for seabed characterization based on impulse responses from various media. Initially, a decision tree architecture is developed, leveraging selected features derived from the impulse responses. Key features include kurtosis, skewness, signal strength, the Hurst Exponent, and characteristics extracted from time-frequency representations of the received signals. Training datasets are constructed by extracting those features from noisy signals, while testing is conducted using a separate dataset along the same manner. The performance of the model is assessed across varying Signal-to-Noise Ratios. Furthermore, multilayer perceptrons are applied to the same datasets for training and testing. The study compares the two machine learning techniques, highlighting their respective advantages and limitations. The findings provide valuable insights into the efficacy of these methods for seabed characterization.
• Corresponding author: Dr Zoi-Heleni Michalopoulou
  Affiliation: Department of Mathematical Sciences, New Jersey Institute of Technology
  Country: United States