A novel nature-inspired feature selection algorithm for efficient moisture estimation in fruits using RF-Sensed data

We propose a novel hybrid nature-inspired feature selection algorithm that unifies update mechanisms from Grey Wolf Optimizer (GWO), Artificial Bee Colony (ABC), and Bat Algorithm (BA). The resulting framework enables optimized machine learning models for precise grape moisture estimation from radio frequency (RF)-sensed data, addressing key challenges in smart agriculture. Performance is assessed in two phases: (1) pairing the feature selection method with a gated recurrent unit (GRU) model and comparing it against benchmark optimizers, and (2) integrating it with a customized convolutional neural network variant (CNN-R) designed for regression. The proposed feature selection technique demonstrates superior performance across all evaluation metrics. When combined with GRU, it achieves significantly lower root mean square error (RMSE) and mean absolute error (MAE) alongside a higher R 2 (best-case: 0.999) compared to benchmark methods. With CNN-R, it maintains equally competitive results, validating its architecture-independent effectiveness. Crucially, the study shows that convolutional neural network (CNN), when adapted through CNN-R, can rival traditional regression models like GRU on numerical data.