ARTIFICIAL NEURAL NETWORK-DRIVEN FORECASTING OF MILLET YIELD BASED ON CLIMATIC INDICATORS IN KANO STATE, NIGERIA

Abstract

Machine learning algorithms like artificial neural networks (ANN) are capable of providing accurate predictions of crop yields based on climatic variables. Meteorological data from 1993 to 2023 and Millet yield from 2010 to 2023 were obtained from the National Aeronautics and Space Administration (NASA) and the Kano Agricultural and Rural Development Agency (KNARDA) respectively. The ANN algorithm was used in predicting Millet yield based on the dataset of the climatic variables. Seven climatic variables (Rainfall, Maximum, temperature, Minimum temperature Relative humidity, Sunshine hours, Wind speed and direction) were used as input neurons for the ANN algorithm. Three hidden neurons were used to train the ANN network in order to understand the pattern of the dataset for better prediction. The maximum prediction in Millet yield was achieved using three hidden neurons. Thus, the best fitting model for the ANN prediction was obtained with an R-squared (R2) of 1 (100%) for the training, and validation with an R2of 0.99 (99%). The study indicated that the model is very reliable and statistically significant in predicting the variability of Millet yield. Therefore, based on the training and validation, it is a perfect linear graph which shows same values for both the actual and predicted variables. Findings from the study reveals that the ANN has the capacity of identifying patterns in historical data and can make prediction of Millet yield variability possible. This study therefore recommends its use in supporting the development of management practices that are best adapted to climate variability and provides an insight on how to reduce risks associated with reduction in crop yield.