Predicting land use change with data-driven models
Predviđanje promena u korišćenju zemljišta primenom modela vođenih podacima (DATA-DRIVEN MODELS)
Doktorand
Samardžić-Petrović, MilevaMentor
Bajat, BranislavČlanovi komisije
Kovačević, MilošCvijetinović, Željko
Dragićević, Suzana
Đorđević, Dejan
Metapodaci
Prikaz svih podataka o disertacijiSažetak
One of the main tasks of data-driven modelling methods is to induce a
representative model of underlying spatial - temporal processes using past data
and data mining and machine learning approach. As relatively new methods,
known to be capable of solving complex nonlinear problems, data-driven methods
are insufficiently researched in the field of land use. The main objective of this
dissertation is to develop a methodology for predictive urban land use change
models using data-driven approach together with evaluation of the performance of
different data-driven methods, which in the stage of finding patterns of land use
changes use three different machine learning techniques: Decision Trees, Neural
Networks and Support Vector Machines. The proposed methodology of data-driven
methods was presented and special attention was paid to different data
representation, data sampling and the selection of attributes by four methods (χ2,
Info Gain, Gain Ratio and Correlation-based Feature Subset) t...hat best describe the
process of land use change. Additionally, a sensitivity analysis of the Support
Vector Machines -based models was performed with regards to attribute selection
and parameter changes. Development and evaluation of the methodology was
performed using data on three Belgrade municipalities (Zemun, New Belgrade and
Surčin), which are represented as 10×10 m grid cells in four different moments in
time (2001, 2003, 2007 and 2010).
The obtained results indicate that the proposed data-driven methodology provides
predictive models which could be successfully used for creation of possible
scenarios of urban land use changes in the future. All three examined machine
learning techniques are suitable for modeling land use change. Accuracy and
performance of models can be improved using proposed balanced data sampling,
including the information about neighbourhood and history in data
representations and relevant attribute selections. Additionally, using selected
subset of attributes resulted in a simple model and with less possibility to be
overfitted with higher values of Support Vector Machines parameters.
Један од главних задатака моделирања метода вођених подацима (Data-driven methods) је проналажење репрезентативног модела испитивног просторно временског процеса, применом података из прошлости и Data Mining и Machine Learning приступа...
Fakultet:
Универзитет у Београду, Грађевински факултетDatum odbrane:
01-10-2014Projekti:
- Uloga i implementacija državnog prostornog plana i regionalnih razvojnih dokumenata u obnovi strateškog istraživanja, mišljenja i upravljanja u Srbiji (RS-MESTD-Integrated and Interdisciplinary Research (IIR or III)-47014)