Publication

Abstract : Multiple Linear Regression (MLR) is the elementary machine learning approach for determining linear relationships among different variables. An improved linear regression method referred as Fixed-Step Validation (FSV) is introduced in this paper for predicting groundwater levels using factors such as rainfall and temperature. The analysis is on 385 observation wells across Kerala in the months January, April, August and November. The model is trained using datasets from 1996-2010 and tested using 2011-2016 datasets. The developed FSV method follows fixations at each stage after training 70% of data. In monthly order, the percentages of improved models generated were 85.2%, 80.8%, 83.6% and 77.9%. The validation performed on data from 2017-2020 resulted in absolute errors in the range 0 to 1 for at least 60% and up to 86.6% of the better modelled wells. The proposed model is found to be better than the benchmark cross-validated MLR technique.