Performance and Robustness Analysis of Advanced Machine Learning Models for Predicting the Required Irrigation Water Amount

Abstract

The agricultural sector plays a pivotal role in ensuring global food security, particularly in light of significant population growth. The demand for food is increasing substantially, while crop production may not sufficiently meet these rising needs. Water scarcity is one of the main problems that poses a significant challenge to the agriculture sector, exacerbated by inefficiencies in traditional irrigation methods. Accurate prediction of plant water requirements is essential to address this issue. This paper proposes advanced machine learning (ML) and deep learning (DL) models to accurately predict the daily water amount (quantity) needs of greenhouse plants using various air and soil data parameters. Various data preprocessing techniques were applied to prepare the data for the proposed models. In addition, due to the different nature of the proposed models, two different data splitting methods were used to split data into inputs and outputs (Simple data preparation for the ML models and time series data preparation for the time series DL models).Results indicate that the Multi-Layer Perceptron (MLP) model consistently outperformed other models, demonstrating superior stability and efficiency across different data optimization phases. Additionally, both ML and Long-Short Term Memory (LSTM) models exhibited strong performance in different data optimization scenarios. Robustness was evaluated through parameter sensitivity analysis, which revealed that ML models were generally more robust than DL models. This robustness is attributed to the limited number of parameters in ML models, which enhances their reliability compared to the more complex DL models. This study ensures the potential of the proposed models to optimize the irrigation practices, thereby addressing water scarcity issues and improving agricultural productivity