Scholarlink

The implementation of Artificial Intelligence (AI) in agriculture has greatly enhanced yield prediction and decision-making. This study introduces a hybrid time-series prediction architecture that uses the statistical prowess of Autoregressive Integrated Moving Average (ARIMA) and the deep learning strength of Long Short-Term Memory (LSTM) neural networks to achieve more accurate crop yield estimates. ARIMA effectively captures seasonal and linear behavior, while LSTM captures nonlinear residual behavior that ARIMA cannot estimate. The hybrid architecture is developed by modeling historical yield trends with ARIMA and modeling the residuals with LSTM. The final prediction is estimated as the summation of the ARIMA and LSTM predictions. In this work, results from agricultural applications using real-world datasets provide confirmation of ARIMA capturing the general trend, while the benefits of hybrid architectures are context dependent based on the variability of residuals and learning capacity of LSTM. Hence, while hybrid architectures demonstrate promise as more advanced forecasting methodologies, their accuracy is hindered by data limitation and residual noise. Future work will seek to improve residual modeling by using richer datasets and applying auxiliary environmental factors such as precipitation, temperature, and soil quality.

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 Residual Learning-Based Hybrid Forecasting Architecture Integrating ARIMA and LSTM for Spatio-Temporal Modeling of Agricultural Yield Prices

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Trijal Ranganathan,
  • Ajay Viswanagaraj,
  • Jeyadheep Velayutham,
  • Sugandha Saxena,
  • M. Lakshmanan,
  • K. Pradeep Kumar

摘要

The implementation of Artificial Intelligence (AI) in agriculture has greatly enhanced yield prediction and decision-making. This study introduces a hybrid time-series prediction architecture that uses the statistical prowess of Autoregressive Integrated Moving Average (ARIMA) and the deep learning strength of Long Short-Term Memory (LSTM) neural networks to achieve more accurate crop yield estimates. ARIMA effectively captures seasonal and linear behavior, while LSTM captures nonlinear residual behavior that ARIMA cannot estimate. The hybrid architecture is developed by modeling historical yield trends with ARIMA and modeling the residuals with LSTM. The final prediction is estimated as the summation of the ARIMA and LSTM predictions. In this work, results from agricultural applications using real-world datasets provide confirmation of ARIMA capturing the general trend, while the benefits of hybrid architectures are context dependent based on the variability of residuals and learning capacity of LSTM. Hence, while hybrid architectures demonstrate promise as more advanced forecasting methodologies, their accuracy is hindered by data limitation and residual noise. Future work will seek to improve residual modeling by using richer datasets and applying auxiliary environmental factors such as precipitation, temperature, and soil quality.