Recent research has indicated that the pupil diameter (PD) in humans varies with their affective states. However, this signal has not been fully investigated for affective sensing purposes in human-computer interaction systems. This may be due to the dominant separate effect of the pupillary light reflex (PLR), which shrinks the pupil when light intensity increases. In this dissertation, an adaptive interference canceller (AIC) system using the H∞ time-varying (HITV) adaptive algorithm was developed to minimize the impact of the PLR on the measured pupil diameter signal. The modified pupil diameter (MPD) signal, obtained from the AIC was expected to reflect primarily the pupillary affective responses (PAR) of the subject. Additional manipulations of the AIC output resulted in a processed MPD (PMPD) signal, from which a classification feature, PMPDmean, was extracted. This feature was used to train and test a support vector machine (SVM), for the identification of stress states in the subject from whom the pupil diameter signal was recorded, achieving an accuracy rate of 77.78%. The advantages of affective recognition through the PD signal were verified by comparatively investigating the classification of stress and relaxation states through features derived from the simultaneously recorded galvanic skin response (GSR) and blood volume pulse (BVP) signals, with and without the PD feature. The discriminating potential of each individual feature extracted from GSR, BVP and PD was studied by analysis of its receiver operating characteristic (ROC) curve. The ROC curve found for the PMPDmean feature encompassed the largest area (0.8546) of all the single-feature ROCs investigated. The encouraging results seen in affective sensing based on pupil diameter monitoring were obtained in spite of intermittent illumination increases purposely introduced during the experiments. Therefore, these results confirmed the benefits of using the AIC implementation with the HITV adaptive algorithm to isolate the PAR and the potential of using PD monitoring to sense the evolving affective states of a computer user.

The society of today relies a lot on the industry and the automation of factory tasks is more prevalent than ever before. However, the machines taking on these tasks require maintenance to continue operating. This maintenance is typically given periodically and can be expensive while sometimes requiring expert knowledge. Thus it would be very beneficial if one could predict when a machine needs maintenance and only employ maintenance as necessary. One method to predict when maintenance is necessary is to collect sensor data from a machine and analyse it for anomalies. Anomalies are usually an indicator of unexpected behaviour and can therefore show when a machine needs maintenance. Due to concerns like privacy and security, it is often not allowed for the data to leave the local system. Hence it is necessary to perform this kind of anomaly detection in an online manner and in an edge environment. This environment imposes limitations on hardware and computational ability. In this thesis we consider four machine learning anomaly detection methods that can learn and detect anomalies in this kind of environment. These methods are LoOP, iForestASD, KitNet and xStream. We first evaluate the four anomaly detectors on the Skoltech Anomaly Benchmark using their suggested metrics as well as the Receiver Operating Characteristic curves. We also perform further evaluation on two data sets provided by the company Gebhardt. The experimental results are promising and indicate that the considered methods perform well at the task of anomaly detection. We finally propose some avenues for future work, such as implementing a dynamically changing anomaly threshold. / Dagens samhälle är väldigt beroende av industrin och automatiseringen av fabriksuppgifter är mer förekommande än någonsin. Dock kräver maskinerna som tar sig an dessa uppgifter underhåll för att forsätta arbeta. Detta underhåll ges typiskt periodvis och kan vara dyrt och samtidigt kräva expertkunskap. Därför skulle det vara väldigt fördelaktigt om det kunde förutsägas när en maskin behövde underhåll och endast göra detta när det är nödvändigt. En metod för att förutse när underhåll krävs är att samla in sensordata från en maskin och analysera det för att hitta anomalier. Anomalier fungerar ofta som en indikator av oväntat beteende, och kan därför visa att en maskin behöver underhåll. På grund av frågor som integritet och säkerhet är det ofta inte tillåtet att datan lämnar det lokala systemet. Därför är det nödvändigt att denna typ av anomalidetektering genomförs sekventiellt allt eftersom datan samlas in, och att detta sker på nätverkskanten. Miljön som detta sker i påtvingar begränsningar på både hårdvara och beräkningsförmåga. I denna avhandling så överväger vi fyra anomalidetektorer som med användning av maskininlärning lär sig och upptäcker anomalier i denna sorts miljö. Dessa metoder är LoOP, iForestASD, KitNet och xStream. Vi analyserar först de fyra anomalidetektorerna genom Skoltech Anomaly Benchmark där vi använder deras föreslagna mått samt ”Receiver Operating Characteristic”-kurvor. Vi genomför även vidare analys på två dataset som vi har tillhandhållit av företaget Gebhardt. De experimentella resultaten är lovande och indikerar att de övervägda metoderna presterar väl när det kommer till detektering av anomalier. Slutligen föreslår vi några idéer som kan utforskas för framtida arbete, som att implementera en tröskel för anomalidetektering som anpassar sig dynamiskt.

Debris-slides are fast-moving landslides that occur in the Appalachian region including the Great Smoky Mountains National Park (GRSM). Various knowledge and data-driven approaches using spatial distribution of the past slides and associated factors could be used to estimate the region’s debris-slide susceptibility. This study developed two debris-slide susceptibility models for GRSM using knowledge-driven and data-driven methods in GIS. Six debris-slide causing factors (slope curvature, elevation, soil texture, land cover, annual rainfall, and bedrock discontinuity), and 256 known debris-slide locations were used in the analysis. Knowledge-driven weighted overlay and data-driven bivariate frequency ratio analyses were performed. Both models are helpful; however, each come with a set of advantages and disadvantages regarding degree of complexity, time-dependency, and experience of the analyst. The susceptibility maps are useful to the planners, developers, and engineers for maintaining the park’s infrastructures and delineating zones for further detailed geotechnical investigation.

Background: In many severely mentally disordered children, the clinical presentation is complicated by comorbid affective and behavioral dysregulation. Recently, a highly heritable behavioral phenotype of simultaneous deviance on the anxious/depressed, attention problems, and aggressive behavior syndrome scales has been identified on the Child Behavior Checklist Dysregulation Profile (CBCL-DP). The aim of the present pilot study was to determine an equivalent to the CBCL-DP using the Strengths and Difficulties Questionnaire (SDQ). Sampling and Methods: We applied stepwise linear discriminant analyses and receiver operating characteristic (ROC) analysis to data from 543 consecutively referred children and adolescents, aged 5–17 years. The CBCL and the SDQ were completed by parents as part of the diagnostic routine. ICD-10 discharge diagnoses were established in consensus conferences. Results: A combination of five SDQ items (SDQ-Dysregulation Profile, SDQ-DP) yielded the best discrimination of children with and without CBCL-DP and classified 81.0% of the subjects correctly leading to an area under the curve of 0.93. The content of the five SDQ-DP items mirrors well the mixed behavioral phenotype of anxious-depressive, aggressive and attention problems captured by the CBCL-DP. SDQ-DP status was highly correlated with CBCL-DP status and was best defined by a SDQ-DP score ≧5. Conclusions: The psychometric properties of the SDQ-DP have been robustly tested and validated. Based on these results, clinicians may use the SDQ-DP as a useful and economical screening measure to improve the assessment, prevention, and treatment of severe dysregulation in childhood and adolescence. Future investigations should study the longitudinal stability, heritability, and genetic associations of this behavioral phenotype. / Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich.

Physiological signals, which are controlled by the autonomic nervous system (ANS), could be used to detect the affective state of computer users and therefore find applications in medicine and engineering. The Pupil Diameter (PD) seems to provide a strong indication of the affective state, as found by previous research, but it has not been investigated fully yet. In this study, new approaches based on monitoring and processing the PD signal for off-line and on-line affective assessment (“relaxation” vs. “stress”) are proposed. Wavelet denoising and Kalman filtering methods are first used to remove abrupt changes in the raw Pupil Diameter (PD) signal. Then three features (PDmean, PDmax and PDWalsh) are extracted from the preprocessed PD signal for the affective state classification. In order to select more relevant and reliable physiological data for further analysis, two types of data selection methods are applied, which are based on the paired t-test and subject self-evaluation, respectively. In addition, five different kinds of the classifiers are implemented on the selected data, which achieve average accuracies up to 86.43% and 87.20%, respectively. Finally, the receiver operating characteristic (ROC) curve is utilized to investigate the discriminating potential of each individual feature by evaluation of the area under the ROC curve, which reaches values above 0.90. For the on-line affective assessment, a hard threshold is implemented first in order to remove the eye blinks from the PD signal and then a moving average window is utilized to obtain the representative value PDr for every one-second time interval of PD. There are three main steps for the on-line affective assessment algorithm, which are preparation, feature-based decision voting and affective determination. The final results show that the accuracies are 72.30% and 73.55% for the data subsets, which were respectively chosen using two types of data selection methods (paired t-test and subject self-evaluation). In order to further analyze the efficiency of affective recognition through the PD signal, the Galvanic Skin Response (GSR) was also monitored and processed. The highest affective assessment classification rate obtained from GSR processing is only 63.57% (based on the off-line processing algorithm). The overall results confirm that the PD signal should be considered as one of the most powerful physiological signals to involve in future automated real-time affective recognition systems, especially for detecting the “relaxation” vs. “stress” states.

The growing need for fast sampling of explosives in high throughput areas has increased the demand for improved technology for the trace detection of illicit compounds. Detection of the volatiles associated with the presence of the illicit compounds offer a different approach for sensitive trace detection of these compounds without increasing the false positive alarm rate. This study evaluated the performance of non-contact sampling and detection systems using statistical analysis through the construction of Receiver Operating Characteristic (ROC) curves in real-world scenarios for the detection of volatiles in the headspace of smokeless powder, used as the model system for generalizing explosives detection. A novel sorbent coated disk coined planar solid phase microextraction (PSPME) was previously used for rapid, non-contact sampling of the headspace containers. The limits of detection for the PSPME coupled to IMS detection was determined to be 0.5-24 ng for vapor sampling of volatile chemical compounds associated with illicit compounds and demonstrated an extraction efficiency of three times greater than other commercially available substrates, retaining >50% of the analyte after 30 minutes sampling of an analyte spike in comparison to a non-detect for the unmodified filters. Both static and dynamic PSPME sampling was used coupled with two ion mobility spectrometer (IMS) detection systems in which 10-500 mg quantities of smokeless powders were detected within 5-10 minutes of static sampling and 1 minute of dynamic sampling time in 1-45 L closed systems, resulting in faster sampling and analysis times in comparison to conventional solid phase microextraction-gas chromatography-mass spectrometry (SPME-GC-MS) analysis. Similar real-world scenarios were sampled in low and high clutter environments with zero false positive rates. Excellent PSPME-IMS detection of the volatile analytes were visualized from the ROC curves, resulting with areas under the curves (AUC) of 0.85-1.0 and 0.81-1.0 for portable and bench-top IMS systems, respectively. Construction of ROC curves were also developed for SPME-GC-MS resulting with AUC of 0.95-1.0, comparable with PSPME-IMS detection. The PSPME-IMS technique provides less false positive results for non-contact vapor sampling, cutting the cost and providing an effective sampling and detection needed in high-throughput scenarios, resulting in similar performance in comparison to well-established techniques with the added advantage of fast detection in the field.

En France, la recherche contre le cancer est un enjeu majeur de santé publique. On estime notamment que le nombre de nouveaux cas de cancer a plus que doublé entre 1980 et 2012. L’hétérogénéité des caractéristiques tumorales, pour un même cancer, impose des défis complexes dans la recherche de traitements efficaces. Dans ce contexte, des espoirs importants sont placés dans la recherche de biomarqueurs prédictifs reflétant les caractéristiques des patients ainsi que de leur tumeur afin d’orienter le choix de la stratégie thérapeutique. Par exemple, pour les cancers colorectaux métastatiques, il est maintenant reconnu que l’ajout de cetuximab (un anti-EGFR) à la chimiothérapie classique (ici le FOLFOX4), n’apporte un bénéfice qu’aux patients dont le gène KRAS est non muté. Le gène KRAS est ici un biomarqueur prédictif binaire, mais de nombreux biomarqueurs sont le résultat d’une quantification ou d’un dosage. L’objectif de cette thèse est dans un premier temps, de quantifier la capacité globale d’un biomarqueur quantitatif à guider le choix du traitement. Après une revue de la littérature, une nouvelle méthode basée sur une extension des courbes ROC est proposée, et comparée aux méthodes existantes. Son principal avantage est d’être non paramétrique, et d’être indépendante de l’efficacité moyenne des traitements. Dans un second temps, lorsqu’un biomarqueur prédictif quantitatif est étudié, la définition d’un seuil de marqueur au-delà duquel la première option de traitement sera préférée, et en-deçà duquel la deuxième option de traitement sera préférée se pose. Une approche reposant sur la définition d’une fonction d’utilité est proposée permettant alors de tenir compte de l’efficacité des traitements ainsi que de leur impact sur la qualité de vie des patients. Une méthode Bayésienne d’estimation de ce seuil optimal est proposée / In France, the cancer research is a major public health issue. The number of new cancer cases nearly doubled between 1980 and 2012. The heterogeneity of the tumor characteristics, for a given cancer, presents a great challenge in the research of new effective treatments. In this context, much hope is placed in the research of predictive (or treatment selection) biomarkers that reflect the patients’ characteristics in order to guide treatment choice. For example, in the metastatic colorectal cancer setting, it is admitted that the addition of cetuximab (an anti-EGFR) to classical chemotherapy (the FOLFOX4), only improve the outcome of patients with KRAS wild-type tumors. In that context, the KRAS gene is a binary treatment selection marker, but plenty of biomarkers result from some quantifications or dosage measurements. The first aim of this thesis is to quantify the global treatment selection ability of a biomarker. After a review of the existing litterature, a method based on an extension of ROC curves is proposed and compared to existing methods. Its main advantage is that it is non-parametric, and that it does not depend on the mean risk of event in each treatment arm. In a second time, when a quantitative treatment selection biomarker is assessed, there is a need to estimate a marker thereshold value above which one treatment is preferred, and below which the other treatment is recommended. An approach that relies on the definition of a utility function is proposed in order to take into account both efficacy and toxicity of treatments when estimating the optimal threshold. A Bayesian method for the estimation of the optimal threshold is proposed

Background: In many severely mentally disordered children, the clinical presentation is complicated by comorbid affective and behavioral dysregulation. Recently, a highly heritable behavioral phenotype of simultaneous deviance on the anxious/depressed, attention problems, and aggressive behavior syndrome scales has been identified on the Child Behavior Checklist Dysregulation Profile (CBCL-DP). The aim of the present pilot study was to determine an equivalent to the CBCL-DP using the Strengths and Difficulties Questionnaire (SDQ). Sampling and Methods: We applied stepwise linear discriminant analyses and receiver operating characteristic (ROC) analysis to data from 543 consecutively referred children and adolescents, aged 5–17 years. The CBCL and the SDQ were completed by parents as part of the diagnostic routine. ICD-10 discharge diagnoses were established in consensus conferences. Results: A combination of five SDQ items (SDQ-Dysregulation Profile, SDQ-DP) yielded the best discrimination of children with and without CBCL-DP and classified 81.0% of the subjects correctly leading to an area under the curve of 0.93. The content of the five SDQ-DP items mirrors well the mixed behavioral phenotype of anxious-depressive, aggressive and attention problems captured by the CBCL-DP. SDQ-DP status was highly correlated with CBCL-DP status and was best defined by a SDQ-DP score ≧5. Conclusions: The psychometric properties of the SDQ-DP have been robustly tested and validated. Based on these results, clinicians may use the SDQ-DP as a useful and economical screening measure to improve the assessment, prevention, and treatment of severe dysregulation in childhood and adolescence. Future investigations should study the longitudinal stability, heritability, and genetic associations of this behavioral phenotype. / Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich.

This work presents a background on the chemical complexity of fire debris analysis, including an ever-present matrix of pyrolysis products as the catalyst that led to the creation of the National Center for Forensic Science's Fire Debris Database. A selection of these 1,000+ casework-relevant ground truth samples was used to create two newly proposed analyst workflows to connect the current method of categorical reporting with evaluative reporting practices reflective of the strength of the evidence. Both workflows use linear sequential unmasking to help mitigate bias, a discrete scoring system for quantification of the analysis, and receiver operating characteristic (ROC) curves to bridge together categorical and probabilistic reporting by indicating the optimum decision threshold the analysts are operating from when they make a decision. Both workflows also allow a machine-learning component to be included in evaluating the evidence and are practical methods for obtaining validated performances for human and machine decisions. The second workflow includes subjective logic, which provides a means of determining the uncertainty inherent to the opinion made by the analyst and the machine learning computational model. ‘Fuzzy categories' and an opinion triangle connect the opinion offered by the analyst given their perceived uncertainty to the ROC curve so a categorical decision can be made. For each workflow, three analysts independently assessed 20 randomly chosen samples from the Fire Debris Database and followed the ASTM E1618-19 standard fire debris analysis method. The resultant area under the ROC curve for each analyst for each workflow was 0.90 or higher, indicating that all were in the very good to excellent range for diagnostic classifiers, as was the machine learning model tested in the second workflow. Recommendations for implementing a performance validation workflow, how repetitive engagement can help the individual analyst and insights on using these for performance validation and training purposes are also included.

Artificial Neural Networks (ANNs) have been comprehensively researched, both from a computer scientific perspective and with regard to their use for predictive modelling in a wide variety of applications including hydrology and the environment. Yet their adoption for live, real-time systems remains on the whole sporadic and experimental. A plausible hypothesis is that this may be at least in part due to their treatment heretofore as “black boxes” that implicitly contain something that is unknown, or even unknowable. It is understandable that many of those responsible for delivering Early Warning Systems (EWS) might not wish to take the risk of implementing solutions perceived as containing unknown elements, despite the computational advantages that ANNs offer. This thesis therefore builds on existing efforts to open the box and develop tools and techniques that visualise, analyse and use ANN weights and biases especially from the viewpoint of neural pathways from inputs to outputs of feedforward networks. In so doing, it aims to demonstrate novel approaches to self-improving predictive model construction for both regression and classification problems. This includes Neural Pathway Strength Feature Selection (NPSFS), which uses ensembles of ANNs trained on differing subsets of data and analysis of the learnt weights to infer degrees of relevance of the input features and so build simplified models with reduced input feature sets. Case studies are carried out for prediction of flooding at multiple nodes in urban drainage networks located in three urban catchments in the UK, which demonstrate rapid, accurate prediction of flooding both for regression and classification. Predictive skill is shown to reduce beyond the time of concentration of each sewer node, when actual rainfall is used as input to the models. Further case studies model and predict statutory bacteria count exceedances for bathing water quality compliance at 5 beaches in Southwest England. An illustrative case study using a forest fires dataset from the UCI machine learning repository is also included. Results from these model ensembles generally exhibit improved performance, when compared with single ANN models. Also ensembles with reduced input feature sets, using NPSFS, demonstrate as good or improved performance when compared with the full feature set models. Conclusions are drawn about a new set of tools and techniques, including NPSFS and visualisation techniques for inspection of ANN weights, the adoption of which it is hoped may lead to improved confidence in the use of ANN for live real-time EWS applications.