Metaheuristic Algorithms to Enhance the Performance of a Feedforward Neural Network in Addressing Missing Hourly Precipitation

Abstract

This research study investigates the implementation of three metaheuristic algorithms, namely, Grey wolf optimizer (GWO), Multi-verse optimizer (MVO), and Moth-flame optimisation (MFO), for coupling with a feedforward neural network (FNN) in addressing missing hourly rainfall observations, while overcoming the limitation of conventional training algorithm of artificial neural network that often traps in local optima. The proposed GWOFNN, MVOFNN, and MFOFNN were compared against the conventional Levenberg Marquardt Feedforward Neural Network (LMFNN) in addressing the artificially introduced missing hourly rainfall records of Kuching Third Mile Station. The findings show that the proposed approaches are superior to LMFNN in predicting the 20% hourly rainfall observations in terms of mean absolute error (MAE) and coefficient of correlation (r). The best performance ANN model is GWOFNN, followed with MVOFNN, MFOFNN and lastly LMFNN.