Today data-driven techniques are becoming more and more essential for solving real world problems, because of the huge amounts of data that is available for analysis. Discovering patterns in the data that lead to better understanding of the data generating process and to useful predictions is one of the important goals that machine learning helps us to achieve. Real-world data sets are often characterized by having large numbers of examples and they are most probably corrupted by noise. The relationship between predictive variables is often highly non-linear. One recent technique that has been developed to address these issues is the Support Vector Machines (SVMs). The SVMs has been developed as robust tool for classification and regression in noisy, complex domains. The two key features of SVMs are generalization theory, which leads to a principled way to choose a decision boundary and kernel functions, which introduce non-linearity in the feature space without explicitly requiring a non-linear algorithm. Support Vector learning algorithm is of generic nature because it can use many different classifiers; therefore it has much wider applicability than other machine learning methods. This Master of Science study examines and addresses following issues of SVMs, their development, optimisation and applicability in Hydroinformatics: i. Understanding and implementation of SVM algorithm. ii. Development of universal Graphical User Interface for SVMs that exhibits features of easy to use navigation system, rules, parameter selection and wizards. iii. Development of V-SVM software (Java applets with visualization functions). iv. Investigation of influence of different kernel functions to model complexity and generalization performance in case of classification and regression. v. Development of Randomisation Utility for data preparation process as part of SVM training procedure. vi. Development of Data Reduction Utility as part of effort to decrease training time of SVM learning algorithm practically without decrease in generalization performance. vii. Development of single image Linux cluster as a methodology to increase performance of SVM learning algorithm through parallelisation of process and significant decrease of training time. viii. Development of the novel practical approach for SVM parameter selection. ix. Application of all developed methodologies, algorithms and software packages to Hydroinformatics problems, in particular in short term hydrologic forecasting. x. Comparison of SVM results with other ML methods. Keywords: Support Vector Machines (SVM), SVM optimisation, SVM parameter tuning, SVM Java applets, RHUL-Delft SVM, Hydroinformatics.
Back to the list of MSc abstracts