A hybrid Random Committee and Multilayer Perceptron Regressor model predicts particle Froude number in auto-washout drainage systems with sedimented beds and identifies volumetric sediment concentration as the most sensitive variable

The work proposes an MLPR-based and RC-MLPR hybrid machine learning approach for PFr prediction and, within one evaluation framework, compares it with other proposed ML models, state-of-the-art ML models, and existing empirical equations, with RC-MLPR performing better. Relative to relying on empirical equations or a single ML model, it combines an ensemble Random Committee strategy with a Multilayer Perceptron Regressor for the multi-dependency PFr prediction problem. Evidence comes from five heterogeneous datasets collected from existing literature covering a wide range of hydraulic and sediment conditions, assessed with several performance measures including the agreement index; the visible text states the conclusion in abstract form without listing specific values.

Sensitivity analysis indicates that volumetric sediment concentration (Csed) is the most sensitive input variable for PFr prediction by the hybrid RC-MLPR model. This adds model-level variable-importance ordering for PFr prediction beyond an overall accuracy comparison. The statement comes from the authors' sensitivity analysis on the same data and RC-MLPR model; the visible text does not report the specific sensitivity metric or values.

The study develops and validates models on five heterogeneous datasets covering a wide range of hydraulic and sediment conditions and states that the supporting data are openly available. Combining multiple heterogeneous sources means evaluation is not confined to a single experimental condition and provides a reusable public data entry point. The data sources and availability statement appear in the abstract and data availability statement (Zenodo link); the visible text does not describe each dataset's size, source experiments, or splitting scheme.

The result applies to auto-washout drainage settings where the particle Froude number must be estimated, particularly design and operation analyses that keep the channel's deposited bed clean; it presumes the input variable system matches the hydraulic and sediment conditions covered by the training data and that variables such as volumetric sediment concentration are available. For users, it offers a reproducible modeling path built on openly available data and points to prioritizing Csed in monitoring and data collection.

The visible text does not give specific values for the performance measures, dataset sizes, or splitting scheme, nor the sensitivity method used, so the magnitude of RC-MLPR's advantage over other models and empirical equations remains unclear; how the model behaves outside the hydraulic and sediment conditions covered by the training data, and its transferability across drainage system geometries and operating conditions, remain open questions.