Scholarlink

<p>This study investigates banana price prediction and market regime characterization using both supervised and unsupervised machine learning techniques. For the forecasting task, deep learning models including CNN, LSTM, GRU, DeepAR, Informer, and FEDformer are evaluated using historical data. The LSTM model achieves the best performance with an R<sup>2</sup> of 0.9659 and the lowest MAE and MAPE, demonstrating its effectiveness in capturing temporal price patterns. In contrast, transformer-based models underperformed in this setting. To uncover underlying market conditions, unsupervised learning methods are applied. Principal Component Analysis (PCA) is used to reduce dimensionality, and K-Means clustering identifies two distinct market regimes—favorable and challenging—characterized by combinations of meteorological, economic, and agricultural factors such as minimum temperature, population density, cultivated area, and export price. Association rule mining further reveals strong relationships between market factors and price levels, such as the link between low prices and high production volumes or low export prices. The integration of predictive modeling with market structure analysis provides actionable insights for stakeholders, including farmers, exporters, and policymakers. For example, the results highlight clear price–supply relationships, where low export prices combined with high production are associated with lower banana prices, while cooler and less humid climatic conditions can coincide with stronger market performance when supported by high export value. The findings highlight the potential of combining deep learning with unsupervised methods to enhance agricultural market intelligence.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A hybrid machine learning framework combining deep forecasting models and unsupervised analysis for cultivated banana price prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月5日

查看全文
  • Chayutpong Manakul,
  • Jessada Sresakoolchai

摘要

This study investigates banana price prediction and market regime characterization using both supervised and unsupervised machine learning techniques. For the forecasting task, deep learning models including CNN, LSTM, GRU, DeepAR, Informer, and FEDformer are evaluated using historical data. The LSTM model achieves the best performance with an R2 of 0.9659 and the lowest MAE and MAPE, demonstrating its effectiveness in capturing temporal price patterns. In contrast, transformer-based models underperformed in this setting. To uncover underlying market conditions, unsupervised learning methods are applied. Principal Component Analysis (PCA) is used to reduce dimensionality, and K-Means clustering identifies two distinct market regimes—favorable and challenging—characterized by combinations of meteorological, economic, and agricultural factors such as minimum temperature, population density, cultivated area, and export price. Association rule mining further reveals strong relationships between market factors and price levels, such as the link between low prices and high production volumes or low export prices. The integration of predictive modeling with market structure analysis provides actionable insights for stakeholders, including farmers, exporters, and policymakers. For example, the results highlight clear price–supply relationships, where low export prices combined with high production are associated with lower banana prices, while cooler and less humid climatic conditions can coincide with stronger market performance when supported by high export value. The findings highlight the potential of combining deep learning with unsupervised methods to enhance agricultural market intelligence.