Stochastic Models and Neural Networks with Prediction Equations: A Comparative Study Using Weather Data of Quetta, Pakistan

We construct stochastic time series models like Auto Regressive Integrated Moving Average (ARIMA), Seasonal ARIMA (SARIMA) and Auto Regressive Moving average (ARMA) to analyze and forecast weather data. The weather parameters are maximum, minimum temperatures and wind speed of five years from January 2012 to December 2016 of Quetta, Pakistan. Daily variations has been taken to forecast data. ARIMA models are used to forecast and predict the equations for monthly data while the SARIMA models were used on seasonal data and it provides better results for short run forecasting.Weibull Distribution (WD) shows better results on wind data as compare to ARIMA. An Artificial Neural Network (ANN) models for prediction of weather parameters are studied and results are found better as compared to the classical statistical method. The experimental results show that ANN gives better predictive values then traditional stochastic modeling techniques due to their ability to deal with non-linear stochastic data.