Background: In recent times, Machine Learning approaches have gained significant traction in modelling non-linearity features in the field of time series forecasting. In the present investigation, the nonstationary, nonlinearity and non-normality features of the jute price series at the three different commodity markets are dealt using the Machine Learning models. 
 
Methods: An attempt has been made to explore efficient Machine Learning (ML) techniques e.g., Time-Delay Neural Network (TDNN) and Support Vector Regression (SVR) for modelling weekly jute prices in different markets of the Coochbehar district (West Bengal). The nonlinearity pattern of the price series is tested by the Brock-Dechert-Scheinkman (BDS) test. 
 
Result: The results of present study show that nonlinearity is present in the jute price series. Accordingly, the TDNN and SVR models have been applied for modelling and forecasting the nonlinearity of the jute prices. Finally, the TDNN(1:1s:1l), TDNN(10:1s:1l) and TDNN (2:1s:1l) models are outperformed SVR models in the Coochbehar, Baxirhat and Tufanganj markets, respectively in terms of minimum RMSE (157.61, 176.32 and 136.03), MAE (94.11, 95.67 and 90.45)&nbsp; and MAPE (1.51, 1.54 and 1.41) criteria. Hence, the TDNN model is regarded as the best optimal model for forecasting the jute prices in the Coochbehar district.

Performance Comparison of Time Delay Neural Network and Support Vector Regression for Forecasting of Jute Prices in Coochbehar District, West Bengal