Internet traffic classification is an important field which several stakeholders are dependent on for a number of different reasons. Internet Service Providers (ISPs) and network operators benefit from knowing what type of traffic that propagates over their network in order to correctly treat different applications. Today Deep Packet Inspection (DPI) and port based classification are two of the more commonly used methods in order to classify Internet traffic. However, both of these techniques fail when the traffic is encrypted. This study explores a third method, classifying Internet traffic by machine learning in which the classification is realized by looking at Internet traffic flow characteristics instead of actual payloads. Machine learning can solve the inherent limitations that DPI and port based classification suffers from. In this study the Internet traffic is divided into two classes of interest: Video and Other. There exist several machine learning methods for classification, and this study focuses on Support Vector Machine (SVM) to classify traffic. Several traffic characteristics are extracted, such as individual payload sizes and the longest consecutive run of payload packets in the downward direction. Several experiments using different approaches are conducted and the achieved results show that overall accuracies above 90% are achievable. / HITS, 4707

In a modern diesel engine, a fuel injector is a vital component. The injectors control the fuel dosing into the combustion chambers. The accuracy in the fuel dosing is very important as inaccuracies have negative effects on engine out emissions and the controllability. Because of this, a diagnosis that can classify the conditions of the injectors with good accuracy is highly desired. A signal that contains information about the injectors condition, is the engine angular velocity. In this thesis, the classification performance of six common machine learning methods is evaluated. The input to the methods is the engine angular velocity. In addition to the classification performance, also the computational cost of the methods, in a deployed state, is analysed. The methods are evaluated on data from a Scania truck that has been run just like any similar commercial vehicle. The six methods evaluated are: logistic regression, kernel logistic regression, linear discriminant analysis, quadratic discriminant analysis, fully connected neural networks and, convolutional neural networks. The results show that the neural networks achieve the best classification performance. Furthermore, the neural networks also achieve the best classification performance from a, in a deployed state, computational cost effectiveness perspective. Results also indicate that the neural networks can avoid false alarms and maintain high sensitivity.

In this thesis, we investigate the usefulness of a group of features in genre classification problems for news. We choose a diverse feature set, covering features related to content and styles of the texts. The features are divided into two groups: semantic and stylistic. More specifically, the semantic features include genre-exclusive words, emotional words and synonyms. The stylistic features include character-level and document-level features. We use three traditional machine learning classification models and one neural network model to evaluate the effects of our features: Support Vector Machine, Complement Naive Bayes, k-Nearest Neighbor, and Convolutional Neural Networks. The results are evaluated by F1 score, precision and recall (both micro- and macro-averaged). We compare the performance of different models to find the optimal feature set for this news genre classification task, and meanwhile seek the most suitable classifier. We show that genre-exclusive words and synonyms are beneficial to the classification task, in that they are the most informative features in the training process. Emotional words have negative effect on the results. We present the best result of 0.97 by macro-average F1 score, precision and recall on the feature set combining the preprocessed dataset and its synonym sets generated based on contexts classified by the Complement Naive Bayes model. We discuss the results achieved from the experiments and the best-performing models, answer the research questions, and provide suggestions for future studies.

Mänskligt igenkännande med användning av sensorer och maskininlärning är ett fält med många praktiska tillämpningar. Det finns några kommersiella produkter som på ett tillförlitligt sätt kan känna igen människor med hjälp av videokameror. Dock ger videokameror ofta en oro för inkräktning i privatlivet, men genom att läsa det relaterade arbetet kan man hävda att i vissa situationer är en videokamera inte nödvändigtvis mer tillförlitlig än billiga, icke-inkräktande sensorer. Att känna igen antalet människor i ett litet studie / kontorsrum är en sådan situation. Även om det har gjorts många framgångsrika studier för igenkänning av människor med olika sensorer och maskininlärningsalgoritmer, kvarstår en fråga om vilken kombination av sensorer och maskininlärningsalgoritmer som är allmänt bättre. Denna avhandling utgår från att testa fem lovande sensorer i kombination med sex olika maskininlärningsalgoritmer för att bestämma vilken kombination som överträffade resten. För att uppnå detta byggdes en arduino prototyp för att samla in och spara läsningarna från alla fem sensorer i en textfil varje sekund. Arduinon, tillsammans med sensorerna, placerades i ett litet studierum på Malmö universitet för att samla data vid två separata tillfällen medan studenterna använde rummet som vanligt. Den insamlade datan användes sedan för att träna och utvärdera fem maskininlärningsklassificerare för var och en av de möjliga kombinationerna av sensorer och maskininlärningsalgoritmer, för både igenkänningsdetektering och igenkänningsantal. I slutet av experimentet konstaterades det att alla algoritmer kunde uppnå en precision på minst 90% med vanligtvis mer än en kombination av sensorer. Den högsta träffsäkerheten som uppnåddes var 97%. / Human recognition with the use of sensors and machine learning is a field with many practical applications. There exists some commercial products that can reliably recognise humans with the use of video cameras. Video cameras often raises a concern about privacy though, by reading the related work one could argue that in some situations a video camera is not necessarily more reliable than low-cost, non-intrusive, ambient sensors. Human occupancy recognition in a small sized study/office room is one such situation. While there has been a lot of successful studies done on human occupancy recognition with various sensors and machine learning algorithms, a question about which combination of sensors and machine learning algorithms is more viable still remains. This thesis sets out to test five promising sensors in combination with six different machine learning algorithms to determine which combination outperformed the rest. To achieve this, an arduino prototype was built to collect and save the readings from all five sensors into a text file every second. The arduino, along with the sensors, was placed in a small study room at Malmö University to collect data on two separate occasions whilst students used the room as they would usually do. The collected data was then used to train and evaluate five machine learning classifier for each of the possible combinations of sensors and machine learning algorithms, for both occupancy detection and occupancy count. At the end of the experiment it was found that all algorithms could achieve an accuracy of at least 90% with usually more than one combination of sensors. The highest hit-rate achieved was 97%.

No description available.

Dialogue act (DA) classification is an important step in the process of developing dialog systems. DA classification is a problem usually solved by supervised machine learning (ML) approaches that all require hand labeled data. Since hand labeling data is a resource-intensive task, many have proposed to focus on unsupervised or semi-supervised ML approaches to solve the problem of DA classification. This master’s thesis explores a novel method for semi-supervised approach to DA classification: K-Means+HMM. The method combines K- Means and Hidden Markov Model (HMM) modeling in addition to abstracting away the words in the utterances to their part-of-speech (POS) tags and the utterances to their cluster labels produced by K-Means prior to HMM training. The focus are the following hypotheses: H1) incorporating context of the utterances leads to better results (HMM is a method specifically used for sequential data and thus incorporates context, while K-Means does not); H2) increasing the number of clusters in K-Means+HMM leads to better results; H3) increasing the number of examples of cluster labels and hand labeled DAs pairs in K-Means+HMM leads to better results (the examples of pairs are used to create the emission probabilities used to define the HMM). One of the conclusions is that K-Means performs better than K-Means+HMM (the result for K-Means measured with one-to-one accuracy is 35.0%, while the result for K-Means+HMM is 31.6%) given 14 clusters and one example pair. However, when the number of examples is increased to 15 the result is 40.5% for K-Means+HMM; the biggest improvement is when the number of examples is increased to 20 resulting in 44% one-to-one accuracy. That is, K-Means+HMM outperforms K-Means provided that a certain number of examples is given. Another conclusion is that the number of examples has a much larger impact on the results - compared to the number of clusters - thus perhaps concluding that the statement “there is no data like labeled data” holds. / Klassificering av dialoghandlingar är ett viktigt steg i processen för utveckling av dialogsystem. Klassificering av dialoghandlingar är ett problem som vanligtvis löses med hjälp av övervakade maskininlärningsmetoder som alla behöver uppmärkt data. Eftersom uppmärkning av data är en resurskrävande uppgift har många föreslagit att fokusera på oövervakade eller delvis övervakade maskininlärningsmetoder för att lösa problemet av klassificering av dialoghandlingar. Denna masteruppsats utforskar en ny delvis övervakad maskininläningsmetod för klassificering av dialoghandlingar: K-Means+HMM. Föru- tom att metoden kombinerar K-Means och Hidden Markiv Model (HMM) modellering, abstraheras orden i yttranden till deras ordklasstaggar och yttranden till deras klusteretiketter som produceras av K-Means före HMM träningen. Projektets fokus är följande tre hypoteser: H1) en intergration av yttrandenas kontext leder till ett bättre resultat (HMM är en metod som används specifikt för sekventiell data och den integrerar således kontexten, medan K-Means gör inte det); H2) ökning av antalet kluster i K- Means+HMM leder till bättre resultat; H3) ökning av antalet exempel av par av klusteretiketter och dialoghandligar uppmärkta för hand i K- Means+HMM leder till bättre resultat (parexemplen används för att skapa emissionssannolikheter som definierar HMM). En av slutsatserna är att K-Means presterar bättre än K-Means+HMM (resultatet för K-means mätt med en-till-en noggrannhet är 35,0%, medan resultatet för K-Means+HMM är 31,6%) givet 14 kluster och ett exempelpar. Däremot, när antalet av exempelpar ökar till 15 ökar resultatet för K-Means+HMM till 40,5%. Den största ökningen är när antalet exempelpar är 20, vilket ger ett resulat på 44% en-till-en noggrannhet. Med andra ord, presterar K-Means+HMM bätre än K-Means då att ett visst antal exempelpar är tillgängligt. En annan slutsats är att antalet av exempelpar har en mycket större effekt på resultaten jämfört med antalet kluster, vilket då möjligtvis leder till slutsatsen att “det finns ingen bättre data än uppmärkt data”.

The ability to navigate through a continuously changing business landscape has been a success factor for Scania to stay a competitive business, when the landscape continues to change. Digitalization has enabled data to be collected from various sources and the ability to embrace the possibilities that come with it and turn it into an advantage is crucial to make sure that Scania is driving the changing industry. Today, Scania is good at collecting and analyzing data but there is room for improvements when it comes to utilizing the data to create data-driven decision-making. This study aims to investigate the possibility of learning more about the users driving behavior through data-driven driving evaluation. This is done with a machine learning approach where a CNN-GRU neural network with an XGBoost classifier is created to classify triaxial acceleration data into normal or aggressive driving behavior. The findings show that this model architecture has a classification accuracy of 87.80 % and the result is discussed with respect to method implementation, quality of data, hyperparameter tuning, and future studies.

Operational deployment of machine learning based classifiers in real-world networks has become an important area of research to support automated real-time quality of service decisions by Internet service providers (ISPs) and more generally, network administrators. As the Internet has evolved, multimedia applications, such as voice over Internet protocol (VoIP), gaming, and video streaming, have become commonplace. These traffic types are sensitive to network perturbations, e.g. jitter and delay. Automated quality of service (QoS) capabilities offer a degree of relief by prioritizing network traffic without human intervention; however, they rely on the integration of real-time traffic classification to identify applications. Accordingly, researchers have begun to explore various techniques to incorporate into real-world networks. One method that shows promise is the use of machine learning techniques trained on sub-flows – a small number of consecutive packets selected from different phases of the full application flow. Generally, research on machine learning classifiers was based on statistics derived from full traffic flows, which can limit their effectiveness (recall and precision) if partial data captures are encountered by the classifier. In real-world networks, partial data captures can be caused by unscheduled restarts/reboots of the classifier or data capture capabilities, network interruptions, or application errors. Research on the use of machine learning algorithms trained on sub-flows to classify VoIP and gaming traffic has shown promise, even when partial data captures are encountered. This research extends that work by applying machine learning algorithms trained on multiple sub-flows to classification of video streaming traffic. Results from this research indicate that sub-flow classifiers have much higher and more consistent recall and precision than full flow classifiers when applied to video traffic. Moreover, the application of ensemble methods, specifically Bagging and adaptive boosting (AdaBoost) further improves recall and precision for sub-flow classifiers. Findings indicate sub-flow classifiers based on AdaBoost in combination with the C4.5 algorithm exhibited the best performance with the most consistent results for classification of video streaming traffic.

De nombreuses données sont enregistrées dans le cadre d'applications variées et leur analyse est un challenge abordé par de nombreuses études. Parmi ces différentes applications, cette thèse est motivée par l'analyse de parcours patients pour mener des études de pharmaco-épidémiologie. La pharmaco-épidémiologie est l'étude des usages et effets de produits de santé au sein de populations définies. Le but est donc d'automatiser ce type d'étude en analysant des données. Parmi les méthodes d'analyses de données, les approches d'extraction de motifs extraient des descriptions de comportements, appelées motifs, caractérisant ces données. L'intérêt principal de telles approches est de donner un aperçu des comportements décrivant les données. Dans cette thèse, nous nous intéressons à l'extraction de motifs temporels discriminants au sein de séquences temporelles, c'est-à-dire une liste d'évènements datés. Les motifs temporels sont des motifs représentant des comportements par leur dimension temporelle. Les motifs discriminants sont des motifs représentant les comportements apparaissant uniquement pour une sous-population bien définie. Alors que les motifs temporels sont essentiels pour décrire des données temporelles et que les motifs discriminants le sont pour décrire des différences de comportement, les motifs temporels discriminants ne sont que peu étudiés. Dans cette thèse, le modèle de chronique discriminante est proposé pour combler le manque d'approches d'extraction de motifs temporels discriminants. Une chronique est un motif temporelle représentable sous forme de graphe dont les nœuds sont des évènements et les arêtes sont des contraintes temporelles numériques. Le modèle de chronique a été choisi pour son expressivité concernant la dimension temporelle. Les chroniques discriminantes sont, de ce fait, les seuls motifs temporels discriminants représentant numériquement l'information temporelle. Les contributions de cette thèse sont : (i) un algorithme d'extraction de chroniques discriminantes (DCM), (ii) l'étude de l'interprétabilité du modèle de chronique au travers de sa généralisation et (iii) l'application de DCM sur des données de pharmaco-épidémiologie. L'algorithme DCM est dédié à l'extraction de chroniques discriminantes et basé sur l'algorithme d'extraction de règles numériques Ripperk . Utiliser Ripperk permet de tirer avantage de son efficacité et de son heuristique incomplète évitant la génération de motifs redondants. La généralisation de cet algorithme permet de remplacer Ripperk par n'importe quel algorithme de machine learning. Les motifs extraits ne sont donc plus forcément des chroniques mais une forme généralisée de celles-ci. Un algorithme de machine learning plus expressif extrait des chroniques généralisées plus expressives mais impacte négativement leur interprétabilité. Le compromis entre ce gain en expressivité, évalué au travers de la précision de classification, et cette perte d'interprétabilité, est comparé pour plusieurs types de chroniques généralisées. L'intérêt des chroniques discriminantes à représenter des comportements et l'efficacité de DCM est validée sur des données réelles et synthétiques dans le contexte de classification à base de motifs. Des chroniques ont finalement été extraites à partir des données de pharmaco-épidémiologie et présentées aux cliniciens. Ces derniers ont validés l'intérêt de celles-ci pour décrire des comportements d'épidémiologie discriminants. / Data are recorded for a wide range of application and their analysis is a great challenge addressed by many studies. Among these applications, this thesis was motivated by analyzing care pathway data to conduct pharmaco-epidemiological studies. Pharmaco-epidemiology is the study of the uses and effects of healthcare products in well defined populations. The goal is then to automate this study by analyzing data. Within the data analysis approaches, pattern mining approaches extract behavior descriptions, called patterns, characterizing the data. Patterns are often easily interpretable and give insights about hidden behaviors described by the data. In this thesis, we are interested in mining discriminant temporal patterns from temporal sequences, i.e. a list of timestamped events. Temporal patterns represent expressively behaviors through their temporal dimension. Discriminant patterns are suitable adapted for representing behaviors occurring specifically in small subsets of a whole population. Surprisingly, if temporal patterns are essential to describe timestamped data and discriminant patterns are crucial to identify alternative behaviors that differ from mainstream, discriminant temporal patterns received little attention up to now. In this thesis, the model of discriminant chronicles is proposed to address the lack of interest in discriminant temporal pattern mining approaches. A chronicle is a temporal pattern representable as a graph whose nodes are events and vertices are numerical temporal constraints. The chronicle model was choosen because of its high expressiveness when dealing with temporal sequences and also by its unique ability to describe numerically the temporal dimension among other discriminant pattern models. The contribution of this thesis, centered on the discriminant chronicle model, is threefold: (i) a discriminant chronicle model mining algorithm (DCM), (ii) the study of the discriminant chronicle model interpretability through its generalization and (iii) the DCM application on a pharmaco-epidemiology case study. The DCM algorithm is an efficient algorithm dedicated to extract discriminant chronicles and based on the Ripperk numerical rule learning algorithm. Using Ripperk allows to take advantage to its efficiency and its incomplete heuristic dedicated to avoid redundant patterns. The DCM generalization allows to swap Ripperk with alternative machine learning algorithms. The extracted patterns are not chronicles but a generalized form of chronicles. More expressive machine learning algorithms extract more expressive generalized chronicles but impact negatively their interpretability. The trade-off between this expressiveness gain, evaluated by classification accuracy, and this interpretability loss, is compared for several types of generalized chronicles. The interest of the discriminant chronicle model and the DCM efficiency is validated on synthetic and real datasets in pattern-based classification context. Finally, chronicles are extracted from a pharmaco-epidemiology dataset and presented to clinicians who validated them to be interesting to describe epidemiological behaviors.

o presente estudo teve por objetivo investigar a influência de duas diferentes resiliências de materiais de amortecimento e de dois tipos de cabedais de calçados esportivos na cinemática e cinética de membro inferior e na percepção do conforto durante a corrida. Também investigamos as possíveis relações entre o conforto percebido e as variáveis biomecânicas capturadas. Para tal, foram avaliados 42 corredores recreacionais adultos, com no mínimo de um ano de experiência em corrida de rua, com mínimo de dois treinos regulares por semana, e com volume de treino semanal superior a 5 km. Foram avaliadas quatro condições de calçados aleatorizadas para cada corredor (material de amortecimento de baixa resiliência e cabedal estruturado, material amortecimento de alta resiliência e cabedal estruturado, material de amortecimento de baixa resiliência e cabedal minimalista, e material amortecimento de alta resiliência e cabedal minimalista). Após avaliação antropométrica e postural do complexo tornozelo/pé, os corredores realizaram corridas em uma pista de 25 metros em laboratório. A avaliação biomecânica foi realizada usando seis câmeras infravermelhas (VICON T-40, Oxford, UK) a 300 Hz, sincronizadas a duas plataformas de força (AMTI BP-600600, Watertown, USA) para aquisição da força reação do solo a 1200 Hz, e palmilhas instrumentas com sensores capacitivos (Pedar X System, Novel, Munique, Alemanha) a 100 Hz. A percepção subjetiva de conforto em cada condição foi avaliada por meio de um questionário de conforto para calçados. As comparações estatísticas entre os calçados foram verificadas por meio de análises de variância (ANOVAs) para medidas repetidas, e correlação de Pearson para verificar as relações entre o conforto e as variáveis biomecânicas (a=O,05). Realizou-se uma análise de Machine Learning para capturar variáveis da série temporal completa das curvas de cinemática e cinética que discriminassem os calçados estudados. Construímos uma matriz de entrada nas dimensões 1080 x 1242 para a análise por Machine learning. Os resultados demonstram que há uma interação entre as condições de cabedal e material de amortecimento que faz com que as comparações de resiliência se comportem de forma distinta para cabedais minimalistas e para cabedais estruturados. Contrariamente ao esperado, para os calçados de cabedal estruturado, as resiliências não foram diferentes entre si, e para o cabedal minimalista, os corredores apresentaram impactos mais altos com o material de baixa resiliência. A estrutura de cabedal influenciou a absorção de impacto, onde o cabedal minimalista apresentou impactos mais altos que o cabedal estruturado. Sobre o conforto, a condição de cabedal minimalista e material de baixa resiliência obteve as piores notas em cinco de nove quesitos do questionário. Em alguns quesitos ele foi o pior avaliado dentre todas as demais condições (como no amortecimento do calcanhar e no conforto geral). O cabedal minimalista recebeu pior avaliação que os cabedais estrutura dos no quesito controle médio-lateral da avaliação de conforto. Observou-se que a correlação entre as variáveis biomecânicas e as variáveis de conforto considerando todos os calçados conjuntamente, apesar de apresentarem valores significativos para algumas associações, foram sempre correlações fracas, abaixo de 30%. Ao se analisar cada condição de calçado isoladamente, em algumas se observou correlação moderada entre as variáveis biomecânicas e o conforto (r >31%, p < O,05), o que não se verificou em outras condições de calçados. Cada calçado gera condições particulares que favorecem ou não a associação entre conforto e repostas biomecânicas. Sobre a análise de Machine Learning, a metodologia foi capaz de diferenciar com sucesso os dois materiais de resiliência diferentes utilizando 200 (16%) variáveis biomecânicas disponíveis com uma precisão de 84,8%, e os dois cabedais com uma precisão de 93,9%. A discriminação da resiliência da entressola resultou em níveis de acurácia mais baixos do que a discriminação dos cabedais de calçados. Em ambos os casos, no entanto, as forças de reação do solo estavam entre as 25 variáveis mais relevantes. As 200 variáveis mais relevantes que discriminaram as duas resiliências estavam distribuídas em curtas janelas de tempo, ao longo de toda série temporal da cinemática e força. Estas janelas corresponderam a padrões individuais de respostas biomecânicas, ou a um grupo de indivíduos que apresentaram as mesmas respostas biomecânicas frente aos diferentes materiais de amortecimento. Como conclusão, destacamos que o cabedal tem maior influência que o material de amortecimento quando se trata da biomecânica da corrida e conforto subjetivo. Nos cabedais estruturados, a resiliência do material da entressola não diferenciou a biomecânica da corrida. A resiliência do material de amortecimento causa efeitos importantes sobre o impacto do calcanhar (menores loading rate, frequência mediana, pico de pressão em retropé) durante a corrida em cabedais com pouca estrutura. Alterações biomecânicas devido à resiliência do material de amortecimento parecem ser dependentes do sujeito, enquanto as relacionadas à estrutura de cabedal parecem ser mais sujeito independente. Sugere-se ter cautela ao afirmar que um calçado mais confortável também gerará respostas positivas biomecânicas, pois as associações entre essas variáveis analisando todos os calçados conjuntamente foram sempre correlações fracas. As correlações moderadas e particulares de cada condição de calçado com determinadas variáveis de conforto nos levam a concluir que os materiais aplicados nos calçado favorecem mais ou menos a percepção de determinada característica de conforto / The aim of this study was to investiga te the influence of two cushioning materiais with different resiliencies and two types of uppers of sportive shoes on kinematics and kinetics of lower limb and on the subjective perception of comfort during running. We also investigated the potential relationship between the perceived comfort and biomechanical variables analyzed. For this purpose, 42 adult recreational runners were evaluated. lhey had at least one year of experience on running, minimum of two regular running workouts per week, and weekly training volume above 5 km. We evaluated four randomized shoes conditions for each athlete (Iow resilience cushioning material and structured upper, high resilience cushioning material and structured upper, low resilience cushioning material and minimalist upper, and high resilience cushioning material and minimalist upper). After anthropometric and postura I assessment of the foot/ankle complex, runners held trials on a 25 meters long indoor track. Biomechanical data were collected by six infrared cameras (VICON l-40, Oxford, UK) at 300 Hz, synchronized with two force platforms (AMll BP-600600, Watertown, USA) at 1200Hz, and in- shoe plantar pressure insoles (Pedar X System, Nove\" Munich, Germany) at 100 Hz. Subjective perception of comfort in each shoe condition was assessed by a questionnaire of footwear comfort. lhe statistical comparisons between the shoes were verified by analysis of variance (ANOVA) for repeated measures and Pearson\'s correlation to verify the relationship between comfort and biomechanical variables (a=0.05). We conducted a Machine Learning analysis to capture variables from the complete kinematics and kinetics time series, which would be able to discriminate the studied footwear. We build an input matrix in the dimensions of 1080 x 1242 for Machine Learning analysis. There was an interaction between the upper structure and the resilience of cushioning material that made comparisons between resiliencies to behave differently for minimal uppers and for structured uppers. Contrary to expectation, for structured uppers, resiliencies were not different from each other, and for the minimal upper, runners had higher impact with the low-resilience material. lhe upper structure influenced the absorption of impact, in which the minimalist upper presented higher impacts than the structured upper. About comfort, minimalist upper condition and low resilience materiais had the worst grades for five of nine questions of the questionnaire. In some questions it was the worst of ali conditions (such as for the comfort in the heel cushioning and overall comfort). lhe minimalist upper received worse assessment than the structured uppers in the question about the mediolateral control. It was observed that the correlation between biomechanical variables and comfort, considering ali shoe conditions together, despite having significant values for some correlations were weak correlations (r <30%, p <0.05). When each shoe condition is analyzed alone, some footwear conditions had moderate correlation between comfort and biomechanical variables (r >31%, p <0.05L although the same behavior was not observed in other shoe conditions. Each shoe represents a specific condition that favor or not the association between comfort and biomechanical responses. On Machine Learning analysis, the method was able to successfully distinguish between the two different resiliencies using 200 (16%) of available biomechanical variables with an accuracy of 84.8%, and between the 2 uppers with an accuracy of 93.9 %. Discrimination of the resiliencies resulted in lower levels of accuracy than the discrimination of shoe uppers. In both cases, however, the ground reaction forces were among the 25 most important features. The 200 most relevant features which discriminate the two resiliencies were distribuited in short time windows along the kinematic and force time series. These windows corresponded to individual biomechanical patterns, or patterns of a group of people with similar behavior. In conclusion, we emphasize that the upper has greater influence than the resilience of cushioning material when it is about biomechanics of running and subjective comfort of the shoes. In structured uppers, the biomechanics did not differenciate the resiliencies of the midsole materiais. The resilience of the cushioning material has important effects on the heel impact (Iower loading rate, median frequency, peak pressure in rearfoot) during running on shoes with little structure on the upper. Biomechanical changes due to the resilience of the cushioning material seems to be dependent on the subject, while related to the upper structure seems to be more independent of the subject. It is suggested to be cautious to affirm that more comfortable footwear will also let to positive biomechanical responses. That is because the correlations between these variables when analyzing ali the footwear together were always weak. Moderate and positive correlations of each shoe condition with some of comfort variables lead us to conclude that the materiais applied on each footwear favors more or less the comfort perception