| Popis: |
Število brezžičnih naprav danes hitro narašča, posledično se viša stopnja nasičenosti radijskega spektra, saj soobstoj številnih tehnologij povzroča motnje v omrežjih. Zmogljivost obstoječih tehnologij lahko povečamo z razvojem natančnejših cenilk kakovosti povezav. V diplomskem delu predlagamo, uvedemo in ovrednotimo nov pristop za razvoj sistema za napovedovanje kakovosti povezav, ki temelji na gradnji in izpeljavi značilk. Po predhodni analizi množic podatkov o paketih Wi-Fi in Sigfox tvorimo nove značilke in izgradimo klasifikacijski model za napovedovanje kakovosti povezave. Predlagani modeli se razlikujejo glede na točnost in popolnost napovedovanja posameznih vrst povezav, predvsem povezav srednje kvalitete. Najboljši model pravilno uvrsti 95% testnih primerov, kar je bistveno izboljšanje v primerjavi s 60% točnostjo večinskega klasifikatorja. The number of wireless devices is increasing rapidly. The wireless spectrum is thus becoming crowded as various technologies co-exist and interfere with each other. One possible way to improve the performance of existing technologies is to develop accurate link quality estimators. In this thesis, we propose, implement and evaluate a novel approach to link quality prediction based on feature engineering. Following a preliminary analysis of a dataset with Wi-Fi packet traces and a dataset with Sigfox packet traces, we developed new features and built a classification model for link quality prediction. The proposed models vary in performance with respect to accuracy and completeness of predicting different types of links, mainly links of intermediate quality. The best proposed model achieved 95% classification accuracy, which is a substantial improvement compared to the 60% accuracy of the majority classifier. |
