Volume 14, Issue 8, August 2023 Edition - IJSER Journal Publication


Publication for Volume 14, Issue 8, August 2023 Edition - IJSER Journal Publication


IJSER Research Group https://www.ijser.org/forum/index.php Register for IJSER Research Forum      IJSER Xplore IJSER Xplore Research Paper Database

Pages   [1]   [2]
 



Deep Learning Based Neural Network Modelling for Cassava Yield Prediction []


Predicting the yield of cassava is necessary to guide decisions towards its availability for the ever-growing number of people depending on it for food. The dynamism and complexity of cassava yield (CY) prediction make it difficult for linear models to produce accurate predictions. Many CY prediction models are based on linear relationships with attendant inability to extract hidden interactions existing among CY features thereby providing insufficient, inefficient and inaccurate information for CY prediction. In consideration of the need to properly analyze the nature, trend, impact and parameter combination for optimal CY prediction, this work proposes a Deep Learning (DL) CY prediction model with capabilities of deciphering hidden and non-linear relationships among CY parameters. A 5-layer (14-6-4-3-1) DL neural network model was designed. The cassava dataset with 2500 samples was collected and used for cassava yield model building and estimation. Hyperbolic transfer function was deployed in the DL hidden layers while Sigmoid transfer function was used in the output layer to produce the least average training and testing errors of 0.0024 and 0.037 respectively. Investigations regarding CY based on the number of cassava cultivars planted per stand revealed that the cultivation of one stem per stand had a higher contributory effect on average CY than cultivation of more stems per stand. The DL model earned 93.60% and 95.00% for accuracy and precision metrics respectively, indicating improved performance. As further work, other variants of DL would be investigated and compared with the proposed DL model with a view to improving CY prediction.




Pages   [1]