A weighted error-minimizer parameter estimation technique for one-inflated positive Poisson distribution

Abstract

An error-minimizing estimator is always preferred in model fittings. However, each error-minimizing estimator minimizes error differently. This paper combines four error-minimizing estimators, which are root mean-squared error, mean absolute error, root mean-squared log error and mean absolute percentage error via a weighted approach. The estimation involves two levels. In the first-level estimation, the estimated parameters are obtained by minimizing error values differently and separately. In the second-level estimation, the resulting estimates from the first-level estimation are combined by either fixed and controlled weights or free and uncontrolled weights. A real crime dataset on the frequency of drunk drivers was considered for demonstration of the technique.