Daily weather forecast in Tiwi, Albay, Philippines using Artificial Neural Network with missing values Imputation
Karla Louissa Marie D. Sobrevilla, Archie G. Quinones, Kharl Vincent S. Lopez, Virna T. Azana · 2016
Daily weather forecasting system promote better resource management and disaster preparedness. This study developed several models following the Multilayer Feedforward Artificial Neural Network (MLFANN) architecture trained with Resilient Propagation (RPROP) algorithm to predict the daily weather forecast in Tiwi, Albay. The data gathered from Advanced Science and Technology Institute (DOST-ASTI) contains missing values; thus, two datasets were generated. The first dataset used Fourier Fit and Multiple Imputation to impute the missing values while the second dataset was generated by removing the missing values. The neural network models were developed based on three criteria: number of hidden neurons according to Kolmogorov's theorem at variable iterations; initial neuron weights (random or calculated); and dataset used. This study attained the optimal model with 10 hidden neurons at 17000 iterations, calculated initial neuron weights and used the dataset with removed missing values. It attained a mean prediction accuracy of 98.96743%.