<![CDATA[Hydroponic cultivation requires precise and adaptive fertility control to maintain optimal plant growth. Conventional rule-based systems operate reactively and often fail to capture the multivariate and temporal dynamics of sensor data. Unlike previous studies that primarily focus on single-parameter forecasting, this study reformulates hydroponic automation as a multi-label actuator prediction problem, aiming to replicate and generalize rule-based control mechanisms using data-driven learning. A comparative analysis of Artificial Neural Network (ANN), one-dimensional Convolutional Neural Network (1D-CNN), and Long Short-Term Memory (LSTM) models was conducted to simultaneously predict six actuator states in an IoT-based hydroponic system. The dataset consists of 1,152 real multivariate time-series samples collected sequentially at 5-minute intervals, comprising six sensor features and six binary actuator labels derived from agronomic standards. Preprocessing includes Gaussian jitter-based augmentation, Z-score normalization, and sliding-window modeling (window size = 5). Data were split chronologically into 80% for training and 20% for testing, with 10% for validation. Results show that LSTM achieved the highest performance (accuracy up to 0.98; F1-score up to 0.95), demonstrating superior temporal modeling capability. Threshold optimization improved minority-actuator detection, enabling reliable, adaptive hydroponic control.]]>
Copyrights © 2026
