PERFORMANCE EVALUATION OF HYBRID FEED FORWARD BACK PROPAGATION NEURAL NETWORK SYSTEM FOR PREDICTION OF RICE PRODUCTION IN CAUVERY RIVER BASIN OF TAMILNADU

Abstract: This paper discuss the use of hybrid Feed Forward Back Propagation Neural Network (FFBPNN) in predicting the area of rice cultivation and rice production in the seven districts of Cauvery River Basin (CRB) in Tamilnadu. The hybrid FFBPNN was already designed and developed by the researcher was put into use for the seven districts of CRB. This paper provides the various best fitting models incorporated with the FFBPNN system. It compares the observed area of rice cultivation with the predicted area of rice cultivation and also observed rice production with the predicted rice production the seven districts of CRB. The Average Relative Error (ARE) between the observed and predicted data from the hybrid system was computed for 3 seasons each having 5 years for 7 districts totaling 210 data items. The computed ARE % was arranged as a frequency distribution and discussed. It was found that 49% of the ARE % computed between the observed and predicted data is having 0 to 10% error and 18.1% of the ARE % are within the class interval of 10 to 20% error. It was found that only 5.2% of ARE % between the observed and predicted data is having more than 100% error. The high error percent was a small portion of the study carried out. It can be reduced if more input data are taken for predictions.

Keywords: Cauvery River Basin, Hybrid FFBPNN, Rice prediction system.