Forecasting with Neural Networks: A comparative study using the data of emergency service

TL;DR

This study compares neural network models with ARIMA for forecasting emergency service data, showing neural networks often outperform ARIMA in out-of-sample predictions despite fitting challenges.

Contribution

It provides a comparative analysis of neural networks and Box-Jenkins methodology for forecasting emergency service data, highlighting neural networks' superior predictive performance.

Findings

Neural networks outperform ARIMA in out-of-sample forecasts.

Fitting neural networks involves challenges like local minima.

Diagnostic checks are used to compare model performance.

Abstract

This is a case study discussing the supervised artificial neural network for the purpose of forecasting with comparison of the Box-Jenkins methodology by using the data of well known emergency service Rescue 1122. We fits a variety of neural network (NN) models and many problems were revealed while fitting the ANNs model to achieve the local minima. Moreover ANNs model is giving much better out of sample forecasts as compare to the ARIMA model. However we use diagnostic checks for the comparison of models.