SOIL MOISTURE PREDICTION USING SHALLOW NEURAL NETWORK

Soil moisture is the key ingredient for the growth as well as survival of the plants. Prediction of moisture in advance will be helpful for the farmers in the field of agriculture. In this paper multiple linear regression, support vector regression and shallow neural network has been used for the advance prediction of soil moisture. Also a new feature i.e. rain has been included in the analysis work for the prediction purpose to visualize the changes in the results. These regression based techniques were applied on three different datasets. The two datasets are collected from the online repositories and the third dataset is prepared by collecting the data using the sensenuts device (wireless sensor network). The predictor used for the evaluation is the MSE (mean squared error) and R2 (co-efficient of determination). The results of shallow neural network with rain as parameter provides MSE and R2 of 0.032 and 0.923 for 1 day ahead, 0.034 and 0.903 for 2 days ahead and 0.111 and 0.739 for 7 days ahead for Braggs farm dataset. For the Kyeamba dataset the MSE and R2 is 0.12 and 0.97590 for 1 day ahead, 0.172 and 0.97585 for 2 days ahead and 0.19 and 0.97581 for 7 days ahead. For the third dataset the MSE and R 2 of 0.12 and 0.98 for 1 day ahead,0.20 and 0.96 for 2 days ahead and 0.20 and 0.95 for 7 days ahead.