Scholarlink

<p>Rubber is the most important economic crop of Kerala, which plays a significant role in the agrarian economy of the state. The present study aimed to develop a forecasting model for the yield of rubber in Kerala, using the productivity data and meteorological data from the past 62 years, from the year 1960–1961 to 2022–2023. The study used both conventional approaches and advanced machine learning techniques for the model development. Different time series models have been developed based on ARIMA, ARIMAX, NNAR, and NNARX methods. Annual rainfall and annual maximum temperature were used as exogenous variables for this study. The models were compared for accuracy and goodness of fit using RMSE and MAPE values. The conventional models, ARIMA and ARIMAX, performed well in model model-building phase, but the prediction accuracy declined in the validation phase. The models NNAR (2,4) and NNAR (2,3) with exogenous variables have performed equally well in terms of accuracy and model fit across both development and testing phases. NNAR (2,4) has been identified as the best model considering its simplicity.</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

Comparison of statistical models and machine learning algorithms in forecasting rubber yield in Kerala

  • 期刊论文

  • 发表日期：

    2025年10月23日

查看全文
  • K. C. Ayyoob,
  • Debasis Bhattacharya,
  • Kader Ali Sarkar,
  • Digvijay Singh Dhakre

摘要

Rubber is the most important economic crop of Kerala, which plays a significant role in the agrarian economy of the state. The present study aimed to develop a forecasting model for the yield of rubber in Kerala, using the productivity data and meteorological data from the past 62 years, from the year 1960–1961 to 2022–2023. The study used both conventional approaches and advanced machine learning techniques for the model development. Different time series models have been developed based on ARIMA, ARIMAX, NNAR, and NNARX methods. Annual rainfall and annual maximum temperature were used as exogenous variables for this study. The models were compared for accuracy and goodness of fit using RMSE and MAPE values. The conventional models, ARIMA and ARIMAX, performed well in model model-building phase, but the prediction accuracy declined in the validation phase. The models NNAR (2,4) and NNAR (2,3) with exogenous variables have performed equally well in terms of accuracy and model fit across both development and testing phases. NNAR (2,4) has been identified as the best model considering its simplicity.