Scholarlink

<p>For high-value perishable goods such as cherries, accurate price forecasting is crucial for the supply chain, stabilizing producer income, informing policy interventions &amp; real-time decision support at various stakeholder levels. This study enquires whether Deep Learning (DL) architectures can be implemented in real time for daily advisory tools while simultaneously surpassing conventional statistical and Machine Learning (ML) models in predicting cherry prices. To tackle this issue, daily cherry price data from five wholesale markets in India were utilized for the period 2012–2024 to evaluate six forecasting methodologies, i.e., Seasonal Auto-regressive Integrated Moving Average (SARIMA), Prophet, Random Forest (RF), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Transformer. The results show that deep learning models are better at capturing non-linear price changes than statistical or tree-based methods. The LSTM and Transformer models consistently performed better across all error metrics (MAE, RMSE, sMAPE, MFE, NMBE, and DA). During the 2025 cherry season, the best-performing LSTM was used as a live, web-based forecasting system that accepted real-time submissions from market officials and produced daily predictions based on field conditions. After the season, we checked the actual prices against the forecasted values and found that they were correct with more than 92 per cent accuracy for market-variety-grade combinations. In major markets like Azadpur, Narwal, and Parimpora, the sMAPE was often below 5–10 per cent, and the error margins (MAE 5–8; RMSE 8–12) were low. Further, the Diebold–Mariano (DM) test showed that deep learning was statistically superior to the baselines. The current study presents a novel and comprehensive operational methodology for advanced Artificial Intelligence (AI) models in the agriculture &amp; allied sectors of India, establishing a scalable framework for incorporating ML/DL into agricultural market intelligence and advisory systems.</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

Deep learning-enabled cherry price forecasting and real-time system deployment across multi-market supply chains in India

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月10日

查看全文
  • F. A. Shaheen,
  • Aqib Gul,
  • Nazir Ganai,
  • Masroor Majid,
  • Mudasir Rashid,
  • Abid Sultan

摘要

For high-value perishable goods such as cherries, accurate price forecasting is crucial for the supply chain, stabilizing producer income, informing policy interventions & real-time decision support at various stakeholder levels. This study enquires whether Deep Learning (DL) architectures can be implemented in real time for daily advisory tools while simultaneously surpassing conventional statistical and Machine Learning (ML) models in predicting cherry prices. To tackle this issue, daily cherry price data from five wholesale markets in India were utilized for the period 2012–2024 to evaluate six forecasting methodologies, i.e., Seasonal Auto-regressive Integrated Moving Average (SARIMA), Prophet, Random Forest (RF), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Transformer. The results show that deep learning models are better at capturing non-linear price changes than statistical or tree-based methods. The LSTM and Transformer models consistently performed better across all error metrics (MAE, RMSE, sMAPE, MFE, NMBE, and DA). During the 2025 cherry season, the best-performing LSTM was used as a live, web-based forecasting system that accepted real-time submissions from market officials and produced daily predictions based on field conditions. After the season, we checked the actual prices against the forecasted values and found that they were correct with more than 92 per cent accuracy for market-variety-grade combinations. In major markets like Azadpur, Narwal, and Parimpora, the sMAPE was often below 5–10 per cent, and the error margins (MAE 5–8; RMSE 8–12) were low. Further, the Diebold–Mariano (DM) test showed that deep learning was statistically superior to the baselines. The current study presents a novel and comprehensive operational methodology for advanced Artificial Intelligence (AI) models in the agriculture & allied sectors of India, establishing a scalable framework for incorporating ML/DL into agricultural market intelligence and advisory systems.