The nonlinear regression trees for retrieving missed data during sea-level measurement

Area of Science:

• Environmental Engineering
• Machine Learning
• Data Science

Background:

• Sea surface displacement (SSD) is vital in environmental engineering.
• SSD measurements are prone to data loss due to instrument failure and unpredictable events.
• Accurate SSD data is essential for environmental monitoring and analysis.

Purpose of the Study:

• To develop an innovative nonlinear regression trees (NRT) technique for retrieving missing SSD data.
• To compare the performance of NRT models based on Support Vector Regression (SVR) and Adaptive Neuro-Fuzzy Inference System (ANFIS).
• To validate the developed models against state-of-the-art algorithms for environmental data imputation.

Main Methods:

• Developed two NRT models: NRT-SVR and NRT-ANFIS.
• Utilized nonlinear machine learning algorithms at the end nodes of regression trees.
• Trained models using pressure and SSD data from Acoustic Doppler Current Profilers (ADCPs) as input parameters.
• Validated model performance against Random Forest (RF) and other existing algorithms.

Main Results:

• NRT methods demonstrated a 71.13% improvement in prediction accuracy over Random Forest, with an average RMSE of 0.019 m.
• The NRT-SVR model achieved the highest accuracy, exhibiting the lowest RMSE (0.009 m) and MAE (0.002 m).
• NRT-SVR also recorded superior performance metrics, including R² (0.997), NSE (0.997), and IA (0.999).

Conclusions:

• The developed NRT models offer a simple yet highly efficient approach for imputing missing SSD data.
• NRT-SVR is identified as the most accurate model for SSD data retrieval.
• The NRT technique shows potential for application in pattern recognition for various environmental and engineering challenges.