Syahputra et al. (2025) A Predictive Model for Crop Irrigation Schedulling Using Machine Learning and IoT-Generated Environmental Data

Identification

• Journal: Journal of Applied Informatics and Computing
• Year: 2025
• Date: 2025-10-06
• Authors: Rizki Syahputra, Dewi Andriani
• DOI: 10.30871/jaic.v9i5.10193

Research Groups

• Industrial Engineering Department, Universitas Teuku Umar
• Agrotechnology Department, Universitas Teuku Umar

Short Summary

This study develops and evaluates a machine learning model for predicting optimal irrigation schedules using real-time environmental data collected from an Internet of Things (IoT) system. The Long Short-Term Memory (LSTM) neural network model achieved high predictive accuracy (R² = 0.95), enabling a shift from reactive monitoring to proactive, data-driven precision agriculture.

Objective

• To develop and evaluate a machine learning model (Long Short-Term Memory neural network) to forecast future soil moisture levels using real-time IoT-generated environmental data, thereby enabling proactive and optimized irrigation recommendations for precision agriculture.

Study Configuration

• Spatial Scale: Universitas Teuku Umar (UTU) farm, Aceh, Indonesia.
• Temporal Scale: Six consecutive days of data collection, with measurements recorded at five specific time intervals (14:00 to 18:00) daily, resulting in 30 synchronized observations.

Methodology and Data

• Models used:
  • Long Short-Term Memory (LSTM) neural network (primary predictive model).
  • Multivariate Linear Regression (MLR) (baseline for comparison).
  • Decision Tree Regression (baseline for comparison).
• Data sources:
  • Real-time environmental data collected from a previously validated IoT smart farming monitoring system.
  • Sensors: DHT11 for ambient temperature and humidity, FC-28 for soil moisture.
  • Microcontroller: ESP32.
  • Data preprocessing: Min-Max scaling for normalization, Savitzky–Golay filter for denoising, data structured for time-series with a 3-step look-back period.
  • Dataset split: First four days (20 time steps) for training, final two days (10 time steps) for testing.

Main Results

• The optimized LSTM model, configured with 100 training epochs and a 3-step look-back period, achieved high predictive accuracy on the test set:
  • Mean Absolute Error (MAE): 2.98%
  • Root Mean Square Error (RMSE): 2.35%
  • R-squared (R2): 0.95
• The LSTM model significantly outperformed baseline models:
  • Multivariate Linear Regression: R² = 0.68, RMSE = 4.52%
  • Decision Tree Regression: R² = 0.74, RMSE = 3.87%
• Residual analysis confirmed the model's robustness, showing a random and uniform scatter of errors without systematic bias.

Contributions

• Successfully transitioned an IoT-based smart farming system from reactive, real-time monitoring to a proactive, intelligent decision-support platform for irrigation scheduling.
• Developed and validated a machine learning model (LSTM) capable of accurately forecasting soil moisture levels using real-time environmental data.
• Demonstrated that the LSTM model significantly outperforms traditional statistical methods, achieving a substantially better model fit and lower error rates.
• Provides a practical pathway towards optimizing irrigation schedules, leading to potential water savings, reduced labor, and enhanced crop health in precision agriculture, particularly relevant for regions relying on traditional farming methods.

Funding

Not explicitly stated in the paper.

Citation

@article{Syahputra2025Predictive,
  author = {Syahputra, Rizki and Andriani, Dewi},
  title = {A Predictive Model for Crop Irrigation Schedulling Using Machine Learning and IoT-Generated Environmental Data},
  journal = {Journal of Applied Informatics and Computing},
  year = {2025},
  doi = {10.30871/jaic.v9i5.10193},
  url = {https://doi.org/10.30871/jaic.v9i5.10193}
}