Priyanto, Muhammad Faisal, Mochamad Imamudin

Artificial Neural Network-Based Forecasting of Rice Yield Using Environmental and Agricultural Data

Introduction

Artificial neural network-based forecasting of rice yield using environmental and agricultural data. Predict high-accuracy rice yield in Indonesia using Artificial Neural Networks (ANN) with environmental and agricultural data. Achieves 98.11% R² to support sustainable practices and food security.

Abstract

This study presents a high-accuracy predictive model for rice production in Indonesia using Artificial Neural Networks (ANN), achieving an R² of 98.11%, Mean Absolute Error (MAE) of 0.0966, and Mean Squared Error (MSE) of 0.0189. Climate variability remains a significant challenge to rice cultivation in regions like Malang City, where unpredictable environmental factors such as rainfall, temperature, and humidity hinder effective crop planning and yield estimation. To address this, we developed a Multilayer Perceptron (MLP)-based ANN model incorporating agro-environmental variables: rainfall, temperature, humidity, harvested area, and production quantity. Historical data from 2009 to 2024 were sourced from the Meteorology, Climatology, and Geophysics Agency (BMKG) and the Central Statistics Agency (BPS). The dataset underwent preprocessing, including cleaning, feature extraction, Z-Score normalization, and partitioning into training and testing sets. The proposed ANN architecture consists of an input layer, three hidden layers, and an output layer for regression tasks. Comparative evaluation against Random Forest, K-Nearest Neighbors, and Support Vector Regression demonstrated the ANN’s superior ability to model complex nonlinear relationships in agricultural data. The results highlight the role of intelligent data-driven systems in enhancing the accuracy of yield forecasting, supporting sustainable agricultural practices, and informing national food security policy.

Review

The study presents a compelling application of Artificial Neural Networks (ANNs) for high-accuracy rice yield forecasting in Indonesia, specifically focusing on Malang City. The reported performance metrics (R² of 98.11%, MAE of 0.0966, MSE of 0.0189) are exceptionally strong, suggesting a robust predictive capability. The authors’ use of a Multilayer Perceptron (MLP) model, integrating diverse agro-environmental variables such as rainfall, temperature, humidity, harvested area, and production quantity, along with comprehensive historical data from reputable sources (BMKG, BPS) spanning 2009-2024, demonstrates a thorough approach to data collection and model development. The comparative analysis against established machine learning models (Random Forest, K-Nearest Neighbors, Support Vector Regression) further underscores the ANN's superior ability to capture complex non-linear relationships, highlighting its potential utility for sustainable agricultural planning and national food security policies. While the methodology outlines essential steps like data preprocessing (cleaning, feature extraction, normalization, partitioning) and a clear ANN architecture (input, three hidden, output layers), certain aspects require further clarification to fully assess the model's validity and generalizability. A key point needing elaboration is the precise definition and usage of "production quantity" as an input variable, especially when the study aims to forecast "rice yield." Yield is typically derived from production quantity and harvested area. If "production quantity" refers to the same time period as the predicted "yield," or is highly correlated with it in a way that implies data leakage, this could artificially inflate performance metrics. Clarification on the temporal granularity of the data (e.g., monthly, yearly) and how "harvested area" and "production quantity" are integrated into the feature set for forecasting the *future* yield is also crucial. Furthermore, while comparative performance is stated as superior, providing the metrics for the other models in the abstract would offer greater context for the claimed superiority. Despite these minor points for clarification, the paper makes a significant contribution to the field of agricultural forecasting, particularly for a critical crop like rice in a vulnerable region. The high predictive accuracy achieved by the ANN model, coupled with its potential to inform crucial agricultural and food security policies, is highly commendable. The study effectively demonstrates the utility of intelligent data-driven systems for navigating climate variability challenges. We recommend that the authors provide further details on the specific handling of the "production quantity" input variable and the exact definition of the predicted output (yield vs. total production) within the main text. Addressing these points would further solidify the paper's strong methodological foundation and enhance its overall impact. Subject to these clarifications, this paper warrants publication.

Full Text

Comments

You need to be logged in to post a comment.