Multi-Time Series Forecasting for Regional Emergency Call Demand

Introduction

Everywhere in the world, road accidents, fires, and crimes pose a constant threat to humans. Emergencies may emanate from incidents resulting in fatalities, bodily harm, or destruction of property. Emergencies have the potential to jeopardize public safety, health, and overall well-being. Furthermore, both natural and anthropogenic disasters can result in a significant number of injuries within a brief period (Hugelius et al., 2020). Emergency services play a pivotal role in healthcare, with demand for them surging significantly in recent years (Oberlin et al., 2020). Emergency Medical Services are crucial for delivering immediate medical care in cases of serious incidents or injuries outside the hospital setting. The presence of EMS stands as a fundamental element within an effective healthcare system due to its vital function in preserving lives and diminishing mortality and morbidity rates, particularly in instances involving grave accidents and acute illnesses (Aringhieri et al., 2017).

Furthermore, the objectives of Emergency Medical Services encompass prompt provision and delivery of care to patients (Nickel et al., 2016), Provision of out-of-hospital care services, coupled with the swift relocation of individuals to suitable healthcare establishments following instances of significant or moderate injuries or emergencies (Jánošíková et al., 2019). In other words, reducing ambulance response time is the primary measure of EMS performance worldwide (Al-Shaqsi, 2010).

Additionally, the positioning of the ambulance is crucial for the efficient functioning of Emergency Medical Services to alleviate distress and minimize injuries from accidents (Brotcorne et al., 2003). Hence, accurately predicting the requisite quantity of ambulances for every region along with their precise placement yields beneficial outcomes and enhances the effectiveness of service provision (Jaldell et al., 2014). Univariate and multivariate are the two primary categories of time series prediction models (Serin et al., 2021). The difference between them lies in the quantity of variables or characteristics utilized in the prediction process. Univariate models concentrate on forecasting future values of a single variable using its historical data, while multivariate models forecast multiple variables simultaneously based on their past values [10],[11].

But in another way, Multi Time Series (No Multivariate Time Series) This refers to a situation where you have multiple distinct time series datasets, each representing a different sequence of data. For example, you might have separate time series for stock prices, customer transactions, and website traffic, all unrelated to each other. These are separate time series, and they are not combined into a single multivariate time series dataset. It's crucial to differentiate multiple time series from multivariate time series. In multivariate analysis, we have a single time series dataset containing multiple variables measured at each time point. These variables are assumed to be interrelated, and the analysis focuses on modeling these relationships to understand the system's dynamics. In contrast, independent multiple time series datasets comprise separate sequences, and the focus of the analysis lies on understanding the individual characteristics and patterns within each series as in Figure. 1.

Figure 1.

The independent multi series (no multivariate)