| Abstract: | Irrigation is a critical method for managing farmland resources such as water and fertilizers. In
Ethiopia, irrigation has been extensively used, and to modernize the current irrigation system in
terms of water management, I have designed a machine learning-based system that automates
water management to enhance irrigation efficiency. This study utilized soil chemical data
collected from farmlands in the Oromia Region, East Showa Zone, Adama Woreda, provided by
the Ethiopian Institute of Agricultural Transformation (EIAT). We collected a total of 90 soil
features using various preprocessing techniques to address issues that could render the data
unusable by machine learning algorithms. Additionally, thresholding and weighted sum analysis
were applied to prepare the data for water management purposes and to generalized decisionmaking.
To develop our classification model for water management, we implemented three machine
learning algorithms: Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and
Gated Recurrent Unit (GRU). These algorithms are capable of handling non-linear issues present
in the data. We employed hyperparameters such as Adam optimizers and activation functions
(sigmoid, tanh, and ReLU), along with evaluation metrics including accuracy, precision, recall,
and F1 score.
By applying these parameters in combination with the three algorithms, developed machine
learning models with accuracy rates of 95.4%, 95.8%, and 94.3% for MLP, LSTM, and GRU,
respectively, after multiple training sessions using various parameter combinations. This study
demonstrates the potential of machine learning models to significantly improve water
management in irrigated farmlands, contributing to the sustainable use of water resources in
agriculture. |
