Forecasting of location occupancy by means of cell phone network data

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2019-10-21

Department

Major/Subject

Data Science

Mcode

SCI3095

Degree programme

Master's Programme in ICT Innovation

Language

en

Pages

71

Series

Abstract

This thesis examines the feasibility of building a forecasting model capable to predict the future location occupancy for different places in the UK. For this matter, historic data of the number of people in each location as well as other data sources during the same period of time have been used. Existing literature has researched the performance of different types of predictive models for time series. Some studies, with datasets from different sources and different time intervals, have been carried out at academia. After these studies, experts cannot agree on one best model and they tend to say that the performance of the models depends on the characteristics of the particular signal to forecast. In this thesis, a set of time series forecasting models have been tested on different time series signals corresponding to different locations. The code implemented provides an interface for the user to select the locations, models, and several other forecasting options in order to perform a proper evaluation of the predictions. Results obtained are not biased from what could be expected from the beginning, and the outcomes obtained with the small amount of data available tend to suggest that there is no single best model for all the different predicted locations. The use of data transformations and external regressors have also been tested in order to improve the performance of the models. Although some combinations of regressors and data transformations seem to improve some of the models, no promising conclusions can be concluded since this improvement does not apply for all the cases.

Description

Supervisor

Solin, Arno

Thesis advisor

Solin, Arno

Keywords

time series, machine learning, data science, forecasting

Other note

Citation