Scenario and Estimation of Walnut Production through Statistical and Artificial Neural Network models

Abstract

newline The Jammu and Kashmir holds an important position in horticultural production and contributes in GSDP (gross state domestic product) upto 7 percent. It is the fundamental strength of the State economy and is a source of livelihood for 33 lac population (Horticulture department, Govt. of JandK). The major walnut production stretches are spread mostly in all the districts of Jammu and Kashmir. Jammu and Kashmir produced 307.11 thousand metric tons of walnut with an area under its cultivation of 86.44 thousand hectares (Economic Survey, 2023-24, Govt. of JandK). This study explores the scenario and estimation of walnut production in Jammu and Kashmir (JandK) using statistical models and Artificial Neural Networks (ANN). It evaluates the dynamics of walnut cultivation, through analysis of primary data and secondary data from official horticulture sources, focusing on area, production, export trends, marketing channels and others. The average area and production of walnuts showed an increase of 8.91 percent and 144.62 percent respectively in Phase II (After implementation of NHM) compared to Phase I (before implementation of national Horticulture mission(NHM) 2005-06). A detailed normality analysis using different tests revealed mixed adherence to normality, while Z-scores affirmed most parameters within the acceptable range. The production trends in Phase I were predominantly influenced by area, while Phase II was driven more by yield changes. Time series models were employed for forecasting, with ARIMA (1,2,2) identified as the best traditional model based on AIC (326.68) and BIC (328.94) criteria. However, ANN models, particularly the Time-Delayed Neural Network H1N3, outperformed ARIMA (1,2,2) in terms of RMSE (79961.52) and R² (0.98) values. ARIMAX (2,2,2) emerged as the most accurate for short-term predictions based on minimum value of AIC (1086.41), BIC (1098.59) and maximum value of R² (0.96), incorporating significant exogenous variables such as area (0.12*) and export (-0.24**). Also, Support Vector r