Repository of colleges and higher education institutions

Show document
A+ | A- | Help | SLO | ENG

Title:Napovedovanje števila dohodnih klicev sistema javne varnosti za klic v sili 112 ob uporabi odprtih podatkov interneta stvari : Doktorska disertacija
Authors:ID Grašič, Valerij (Author)
ID Boshkoska, Biljana Mileva (Mentor) More about this mentor... New window
Files:.pdf RAZ_Grasic_Valerij_i2021.pdf (5,56 MB)
MD5: 27703F864752AA3F36911A33678DC794
 
Language:Slovenian
Work type:Doctoral dissertation
Typology:2.08 - Doctoral Dissertation
Organization:FIŠ - Faculty of Information Studies in Novo mesto
Abstract:Na svetu je veliko naravnih razmer, kot so potresi, cunamiji, poplave in žled, ki povzročajo nesreče večjega obsega. Informacije o takšnih dogodkih zbirajo sistemi javne varnosti za klic v sili 112. Ključno vprašanje v okviru podane naloge je vnaprejšnja klasifikacija dohodnih klicev na klic v sili 112. Medtem ko napovedovanje dohodnih klicev v Sloveniji temelji na povprečnem in največjem številu dohodnih klicev, je vsebina doktorske disertacije povezana z zagotavljanjem bolj dinamičnega, inteligentnejšega in na umetni inteligenci utemeljenega napovedovanja števila dohodnih klicev na sistem javne varnosti, ki je ovrednoteno za Ljubljano in Slovenijo z upoštevanjem vseh dohodnih klicev na sistem za klic v sili 112. Narejena je primerjava petih različnih metod klasifikacije za mesto Ljubljana in celotno Slovenijo. Skupaj je uporabljenih 20 atributov za Ljubljano ter 176 atributov za Slovenijo. Število dohodnih klicev se razdeli v dva velikostna razreda, to sta razreda regularni in alarm, ter v štiri velikostne razrede, kjer se razred regularni dodatno razdeli še na razrede majhen, normalen in povečan. Podatki so zbrani na dnevni osnovi za dve časovni obdobji, označeni kot prvo (za leta 2013–2016) in drugo (za leto 2018). Za klasifikacijo so uporabljene metode Naive Bayes, SVM, AdaBoostM1, J48 ter Random Forest, in sicer po kvartalih ter za celotno opazovano obdobje. Rezultati ovrednotenja kažejo, da je najboljša metoda Random Forest, dobre rezultate pa izkazujejo tudi metode J48, Naive Bayes, AdaBoostM1 in SVM. V najboljšem klasifikacijskem primeru s podatki za Slovenijo in metodo Random Forest z dvema razredoma je bila dosežena točnost 94,6 % za celotno obdobje in 98,1 % po kvartalih ter za štiri razrede dosežena točnost 69,2 % za celotno obdobje in 82,5 % po kvartalih. Na osnovi rezultatov ovrednotenja so podani dodatni predlogi, kako pristopiti k napovedovanju števila dohodnih klicev za različne primere. Podani rezultati in predlogi v okviru doktorskega dela so korak naprej v smeri napovedovanja dohodnih klicev. S tem je možno izboljšati zavedanje situacije v kontrolnih sobah, kar vključuje tako dinamiko klicev kot tudi vnaprejšnjo pripravljenost različnih služb na izjemne dogodke.
Keywords:javna varnost, 112, pametno mesto, varno mesto, klasifikacija, internet stvari (IoT)
Place of publishing:Novo mesto
Place of performance:Novo mesto
Publisher:{V. Grašič}
Year of publishing:2021
Year of performance:2021
Number of pages:XXXI, str. 317
PID:20.500.12556/ReVIS-7758 New window
COBISS.SI-ID:67602435 New window
UDC:004:351.78(043.3)
Publication date in ReVIS:18.06.2021
Views:2414
Downloads:144
Metadata:XML DC-XML DC-RDF
:
Copy citation
  
Share:Bookmark and Share


Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Licences

License:CC BY-NC-ND 4.0, Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Link:http://creativecommons.org/licenses/by-nc-nd/4.0/
Description:The most restrictive Creative Commons license. This only allows people to download and share the work for no commercial gain and for no other purposes.
Licensing start date:18.06.2021

Secondary language

Language:English
Abstract:Many natural conditions in the world, such as earthquakes, tsunamis, floods and sleet, cause major disasters. The emergency telephone number 112 collects information on such events. A key issue in this doctoral dissertation is the forecasting and classification of incoming 112 emergency calls, in Ljubljana and Slovenia. While the forecasting of incoming calls in Slovenia is based on an average and a maximum number of incoming calls, this dissertation deals with a more dynamic, artificial intelligence-based forecasting of the number of incoming calls. A comparison of five different classification methods is made, for Ljubljana city and the whole country of Slovenia. A total of 20 attributes are used for Ljubljana and 176 attributes for Slovenia. The number of incoming calls is divided into two or four classes; in the case of two classes, these are the regular and alarm classes, while for the four classes, the regular class is further divided into small, normal and increased classes. Data are collected daily for two periods, the first (for 2013–2016) and the second (for 2018). Naive Bayes, SVM, AdaBoostM1, J48 and Random Forest methods are used for the classification, both for individual quarters of the year and the whole period. The results show that the best method is Random Forest, with methods J48, Naive Bayes, AdaBoostM1 and SVM showing good results. For the best classification case, with data for Slovenia and the Random Forest method with two classes, the accuracy was 94.6 % for the whole period and 98.1 % by quarters, and for four classes the accuracy was 69.2 % for the whole period and 82.5 % by quarters. Such promising results and proposals are a step forward in the direction of forecasting incoming calls, which makes it possible to improve the awareness of the situation in the control rooms, including both the dynamics of the calls themselves, as well as the fact that various services prepare in advance for natural disasters or exceptional occurrences.
Keywords:public safety, 112, smart city, safe city, classification, Internet of Things (IoT)


Back