A Study on Logistic Regression Modeling for Classification in Agriculture

Abstract

Classification and prediction in agricultural systems are quite useful for planning purposes. In this study, logistic regression modeling has been employed for classification purposes in the field of agriculture. The data pertain to the area of agricultural ergonomics with dependent variable as the presence or absence of discomfort for the farm labourers in operating farm machineries along with associated quantitative and qualitative regressors. From the different possible variable subset datasets, only appropriate logistic regression models that best fit these datasets have been selected for further study. Relevant goodness of fit and predictive ability measures has been utilized for evaluating the fitted models. A single best regressor i.e. load given to the farm machinery during operation has also been identified by employing variable selection based on collinearity diagnostics and stepwise logistic regression. Comparison made between the length of confidence intervals of estimates from Maximum Likelihood Estimation (MLE) and quadratic bootstrap methods upon the original sample using the single best regressor revealed that the latter is better than the former as quadratic bootstrap estimates had smaller length of confidence intervals. In addition, resampling based estimation method viz. Quadratic Bootstrap has been applied for estimating the unknown parameters in a simple logistic regression model under a simulation study whose parameter estimates had less bias than that obtained using the conventional MLE procedure without increase in their corresponding estimated variances. Also when the classificatory performances of the logistic regression models (using the best regressor) fitted using both MLE and Quadratic Bootstrap approaches are compared, the results came out to be at par under the two approaches. Classifications of the hold-out datasets revealed that results obtained using logistic regression models were found to be better when compared to those obtained from discriminant function analysis method. Moreover, when comparisons were made among the MLE based logistic regression models, the model with the single best regressor came out to be the best. The study revealed that logistic regression modeling can be employed as a viable alternative for classification purposes in the field of agriculture.