RT Dissertation/Thesis
T1 Machine learning algorithms for pattern visualization in classification tasks and for automatic indoor temperature prediction
A1 Alawadi, Sadi
K1 dimensionality reduction
K1 classification
K1 regression
K1 temperature forecasting
AB This thesis explores aspects in the field of machine learning, and specificallyof pattern classification and regression or function approximation. Althoughthere are many methods of classification for multi-dimensional patterns, ingeneral, they all behave like "black boxes" where the explanation of theiroperation is difficult or impossible. This thesis develops methods of reducingthe dimensionality of data to project multi-dimensional classification problemsover a two-dimensional space (a plane). The classifiers can thus be used tolearn the projected data and to create two-dimensional maps of classificationproblems whose graphic nature makes intuitive and easy to understand, helping toexplain the classification problem. After a review of the existing techniquesfor dimensionality reduction, several methods are proposed to project the multidimensionaldata on the plane, minimizing the overlap between classes. Thesemethods allow to project new patterns not used during the projection learningprocess. Eight types of linear, quadratic and polynomial projections areproposed and combined with four overlapping measures between classes. Theseprojections are compared with another 34 dimensionality reduction methodsexisting in the literature on a wide collection of 71 benchmark classificationproblems. The best results have been obtained by the Polynomial KernelDiscriminant Analysis of degree 2 (PKDA2), which creates visual and selfexplanatorymaps of the classification problems on which a reference classifier(the support vector machine, or SVM) fails only slightly less than on theoriginal multi-dimensional data. A web interface and a local standaloneapplication are also provided, developed using the PHP and Matlab programminglanguages, respectively, which allow to apply these projections in order tovisualize the 2D maps of any classification problem.In the scope of regression, a wide collection of regressors has been applied forthe automatic prediction of temperatures in air conditioning systems (HVAC).These systems have a direct impact on both energy consumption and the comfort ofbuildings, so an accurate and reliable modelling of the temperature behaviorconstitutes the starting point for the development of energy efficiency plans.The use of regressors to predict the evolution of indoor temperature ofbuildings based on internal and external (climatic) conditions allows toevaluate the impact of the modifications in the HVAC systems from a comfortperspective. With the aim of developing an efficient model for HVAC systems,this thesis has evaluated 40 regressors, which belong to 20 different regressorfamilies, using real data generated by an intelligent building, namely theCentro Singular de Investigación en Tecnoloxías da Información (CiTIUS) of theUniversity of Santiago de Compostela (USC). In addition, different models basedon neural networks which allow automatic re-training and on-line learning of newdata have been developed and compared to the previous 20 off-line regressors.The ability of on-line learning provides robustness to the neural models andallows them to: 1) face circumstances never seen in training due to exceptionalclimatic situations; and 2) support alterations in the components of the systemsproduced by errors or changes in the sensor systems.
YR 2018
FD 2018
LK http://hdl.handle.net/10347/16633
UL http://hdl.handle.net/10347/16633
LA eng
DS Minerva
RD 30 abr 2026