No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

Quantitative Structure Activity Relationship (QSAR) is a very useful computa-tional method which has facilitated great progress in drug development [1]. Thismethod can be used to predict a molecule’s activity against a certain target justby comparing its structural characteristics (i.e., molecular descriptors) with thosebelonging to molecules of known activity. QSAR modeling is fueled by online freedatabases consisting of millions of active and inactive molecules and by MachineLearning (ML) Methods that enable data analysis. To ensure successful implemen-tation of ML models, there is a range of evaluation methods to estimate their perfor-mance and applicability domain. So far, a great deal of research has focused on theuse of Support Vector Machines (SVMs) to classify molecules with the use of theirMolecular Signature Fingerprints as descriptors [2]. However, another MachineLearning algorithm, Deep Neural Networks (DNNs), an improvement of single-layer Neural Networks, is rising in popularity in various fields including moleculeclassification. The two models were compared using CPSign software which intro-duces Conformal Prediction, to evaluate the reliability of model predictions basedon performance for individual compounds rather than mean performance on agiven test set. Three types of descriptors were used: Molecular Signature Finger-prints, Extended Connectivity Fingerprints and physicochemical descriptors. Thecomparison showed that Multilayer Perceptron (MLP) which was used as a DNNrepresentative in current context, had performance similar to the shallower SVMmodels but additionally demanded longer training times [3]. It can be concludedthat in the field of QSAR with the aforementioned descriptors, when the numberof examples used for training is not immense, Support Vector Machines might per-form equally well and demand less resources and time than the more sophisticated MLPs.

This master thesis investigated how well a multiclass support vector machine approach is at classifying a fixed number of interpersonal relations between extracted entities of people from Swedish verdicts. With the help of manually tagged extracted pairs of people entities called relations, a multiclass support vector machine was used to train and test the performance of the classification. Different features and parameters were tested to optimize the method, and for the final experiment, a micro precision and recall of 91.75% were found. For macro precision and recall, the result was 73.29% and 69.29% respectively. This resulted in an macro F score of 71.23% and micro F score of 91.75%. The results showed that the method worked for a few of the relation classes, but more balanced data would have been needed to answer the research question to a full extent. / Detta examensarbete utforskade hur bra en multiklass stödvektor- maskin är på att klassificera sociala relationer mellan extraherade personentiteter ur svenska domar. Med hjälp av manuellt taggade par av personentiteter kallade relationer, har en multiklass stödvektormaskin tränats och testats på att klassifiera dessa relationer. Olika attribut och parametrar har testats för att optimera metoden, och för det slutgiltiga exprimentet har ett resultat på 91.75% för båda mikro precision och återkallning beräknats. För makro precision och återkallning har ett resultat på 73.29% respektive 69.29% beräknats. Detta resulterade i ett makro F värde på 71.23% och ett mikro F värde på 91.75%. Resultaten visade att metoden fungerade för några av relationsklasserna men mer balanserat data skulle ha behövts för att forskningsfrågan skulle kunna besvara helt.

Early assessment of tumour response has lately acquired big interest in the medical field, given the possibility to modify treatments during their delivery. Radiomics aims to quantitatively describe images in radiology by automatically extracting a large number of image features. In this context, PET/CT (Positron Emission Tomography/Computed Tomography) images are of great interest since they encode functional and anatomical information, respectively. In order to assess the patients' responses from many image features appropriate methods should be applied. Machine learning offers different procedures that can deal with this, possibly high dimensional, problem. The main objective of this work was to develop a method to classify lung cancer patients as responding or not to chemoradiation treatment, relying on repeated PET/CT images. Patients were divided in two groups, based on the type of chemoradiation treatment they underwent (sequential or concurrent radiation therapy with respect to chemotherapy), but image features were extracted using the same procedure. Support vector machines performed classification using features from the Radiomics field, mostly describing tumour texture, or from handcrafted features, which described image intensity changes as a function of tumour depth. Classification performance was described by the area under the curve (AUC) of ROC (Receiving Operator Characteristic) curves after leave-one-out cross-validation. For sequential patients, 0.98 was the best AUC obtained, while for concurrent patients 0.93 was the best one. Handcrafted features were comparable to those from Radiomics and from previous studies, as for classification results. Also, features from PET alone and CT alone were found to be suitable for the task, entailing a performance better than random.

<p>Most existing classification algorithms either consider all features as equally important (equal weights), or do not analyze consistency of weights assigned to features. When features are not equally important, assigning consistent weights is a not obvious task. In general we have two cases. The first case assumes that a given sample of data does not contain any clue about the importance of features, so the weights are provided by a pool of experts and they are usually inconsistent. The second case assumes that the given sample contains some information about features importance, hence we can derive the weights directly from the sample. In this thesis we deal with both cases. Pairwise Comparisons and Weighted Support Vector Machines are used for the first case. For the second case a new approach based on the observation that the feature importance could be determined by the discrimination power of features has been proposed. For the first case, we start with pairwise comparisons to rank the importance of features, then we use distance-based inconsistency reduction to refine the weights assessment and make comparisons more precise. As the next step we calculate the weights through the fully-consistent or almost consistent pairwise comparison tables. For the second case, a novel concept of feature domain overlappings has been introduced. It can measure the feature discrimination power. This model is based on the assumption that less overlapping means more discrimination ability, and produces weights characterizing the importance of particular features. For both cases Weighted Support Vector Machines are used to classify the data. Both methods have been tested using two benchmark data sets, Iris and Vertebal.</p> <p>The results were especially superior to those obtained without weights.</p> / Master of Computer Science (MCS)