Integrated Model of Demand for Telephone Services in Terms of Microeconometrics

Abstract

The paper presents the results of the testing effectiveness of the integrated model in the short-term forecasting of demand for telephone services in 24-hour cycles. The linear regression model with dichotomous (binary) independent variables was integrated with the feed forward neural network. The regression model was used as a filter of modelled variability of the demand. The neural network was used to model residual variability. The research shows that the integrated model has a higher possibility of approximation and prediction in comparison to the non-integrated linear regression model. The research study was based on data provided by the selected telecommunications network operator. The range of empirical material consisted of hourly counted seconds of outgoing calls and generated by network subscribers in various analytical sections.

Econometric modelling of compound cyclicality of using telecommunication services

“…The obtained goodness-of-fit of the models was definitely higher in comparison to the multi-sectional non-integrated regression model (Kaczmarczyk, 2016). In turn, the results were worse compared to the single-sectional non-integrated regression model (Kaczmarczyk, 2018).…”

“…In the literature of modelling and short-term and mid-term forecasting of the hourly demand for telephone services, the models dedicated to such approximation and prediction are called integrated models (Kaczmarczyk, 2016). The idea of the integrated model consists in the fusion of two different techniques of modelling and forecasting in terms of the analysed demand.…”

Feedforward Neural Networks and the Forecasting of Multi-Sectional Demand for Telecom Services: A Comparative Study of Effectiveness for Hourly Data

“…The modelled and forecasted demand (response variable Y) was hourly counted seconds of outgoing calls within the framework of several different analytical sections. From this, the constructed models (the neural model and the regressive-neural model) can be considered as multi-sectional models [Kaczmarczyk 2016[Kaczmarczyk , 2017. In order to identify the analytical sections, classification factors were specified.…”

“…Under Masters' approach, a regression model or other technique should be used to prepare data for a neural network model. This combination of both models is called a regressive-neural model or integrated model [Kaczmarczyk 2006[Kaczmarczyk , 2016.…”

Microeconometric analysis of telecommunication services market with the use of SARIMA models

“…Therefore, it is not necessary to include them in the prognostic process. When a model is created for short and medium forecast horizon, the following factors should be considered: the type of day (typical working day, Saturday, Sunday, high days, and holidays) hour of the day, category of connection, the type of subscribers, promotions (Kaczmarczyk, 2016).…”