Abstract:
Agriculture plays an important role in Myanmar’s economy by supporting
national food security, employment, and the livelihoods of a large proportion of the
population. However, agricultural productivity is increasingly affected by climate
change, irregular rainfall, rising temperatures, and declining soil quality. These
challenges make it difficult for farmers to select suitable crops for different regions.
Therefore, an intelligent crop recommendation system is needed to help farmers
identify the most suitable crops based on local environmental and agricultural
conditions.
The main objective of this research is to develop an intelligent crop
recommendation system for region-specific agriculture in Myanmar by integrating
Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory
(BiLSTM), Transformer Attention, and Ensemble Boosting techniques. The proposed
system combines multiple sources of agricultural data, including satellite imagery,
vegetation indices such as the Normalized Difference Vegetation Index (NDVI),
climatic variables, soil properties, topographical information, and historical crop
cultivation records. These data are preprocessed through data cleaning, normalization,
and feature extraction to improve data quality and support effective model training.
Within the proposed framework, CNN is responsible for learning spatial
characteristics from satellite images, whereas BiLSTM analyzes time-dependent
patterns in sequential climate and environmental observations. Transformer Attention
is applied to determine the relative importance of different features and their
contributions to crop suitability. In addition, Ensemble Boosting combines predictive
models to enhance classification performance and improve the reliability of the
resulting predictions.
The effectiveness of the proposed hybrid architecture is examined through
comparisons with learning CNN–BiLSTM attention mechanism. Model performance
is assessed using several evaluation measures, namely Accuracy, Precision, Recall, F1-
Score, Area Under the Receiver Operating Characteristic Curve (AUC), Mean Absolute
Error (MAE), and Root Mean Square Error (RMSE).
The primary objective of this research is to develop a dependable crop
recommendation framework that addresses the agricultural requirements of different regions in Myanmar. By identifying crops suitable for local environmental conditions,
the system aims to assist farmers in making informed cultivation choices, increasing
productivity, minimizing the inefficient use of agricultural resources, and adapting to
climate-related challenges. It may also provide useful decision-support information for
agricultural institutions and policymakers involved in sustainable farming
development.
Future improvements may include the incorporation of Internet of Things (IoT)
technologies, Explainable Artificial Intelligence (XAI), and continuously updated
satellite observations. These extensions could expand the framework's capabilities and
support the development of more intelligent, adaptive, and sustainable agricultural
decision-support systems in Myanmar.