This thesis is concerned with developing methodologies that enable existing models to be effectively reused. Results of this thesis are presented in the framework of Quantitative Structural-Activity Relationship (QSAR) models, but their application is much more general. QSAR models relate chemical structures with their biological, chemical or environmental activity. There are many applications that offer an environment to build and store predictive models. Unfortunately, they do not provide advanced functionalities that allow for efficient model selection and for interpretation of model predictions for new data. This thesis aims to address these issues and proposes methodologies for dealing with three research problems: model governance (management), model identification (selection), and interpretation of model predictions. The combination of these methodologies can be employed to build more efficient systems for model reuse in QSAR modelling and other areas. The first part of this study investigates toxicity data and model formats and reviews some of the existing toxicity systems in the context of model development and reuse. Based on the findings of this review and the principles of data governance, a novel concept of model governance is defined. Model governance comprises model representation and model governance processes. These processes are designed and presented in the context of model management. As an application, minimum information requirements and an XML representation for QSAR models are proposed. Once a collection of validated, accepted and well annotated models is available within a model governance framework, they can be applied for new data. It may happen that there is more than one model available for the same endpoint. Which one to chose? The second part of this thesis proposes a theoretical framework and algorithms that enable automated identification of the most reliable model for new data from the collection of existing models. The main idea is based on partitioning of the search space into groups and assigning a single model to each group. The construction of this partitioning is difficult because it is a bi-criteria problem. The main contribution in this part is the application of Pareto points for the search space partition. The proposed methodology is applied to three endpoints in chemoinformatics and predictive toxicology. After having identified a model for the new data, we would like to know how the model obtained its prediction and how trustworthy it is. An interpretation of model predictions is straightforward for linear models thanks to the availability of model parameters and their statistical significance. For non linear models this information can be hidden inside the model structure. This thesis proposes an approach for interpretation of a random forest classification model. This approach allows for the determination of the influence (called feature contribution) of each variable on the model prediction for an individual data. In this part, there are three methods proposed that allow analysis of feature contributions. Such analysis might lead to the discovery of new patterns that represent a standard behaviour of the model and allow additional assessment of the model reliability for new data. The application of these methods to two standard benchmark datasets from the UCI machine learning repository shows a great potential of this methodology. The algorithm for calculating feature contributions has been implemented and is available as an R package called rfFC. / BBSRC and Syngenta (International Research Centre at Jealott’s Hill, Bracknell, UK).

No description available.

This study identified marginal agricultural lands in the Lower Mississippi Alluvial Valley using crop yield predicting models. The Random Forest Regression (RFR) and Multiple Linear Regression (MLR) models were trained and validated using county-level crop yield data, climate data, soil properties, and Normalized Difference Vegetation Index (NDVI). The RFR model outperformed MLR model in estimating soybean and corn yields, with an index of agreement (d) of 0.98 and 0.96, Nash-Sutcliffe model efficiency (NSE) of 0.88 and 0.93, and root mean square error (RMSE) of 9.34% and 5.84%, respectively. Marginal agricultural lands were estimated to 26,366 hectares using cost and sales price in 2021 while they were estimated to 623,566 hectares using average cost and sales price from 2016 to 2021. The results provide valuable information for land use planners and farmers to update field crops and plan alternative land uses that can generate higher returns while conserving these marginal lands.

No description available.

Smart homes are custom-fitted systems for users to manage their home environments. Smart homes consist of devices which has the possibility to communicate between each other. In a smart home system, the communication is used by a central control unit to manage the environment and the devices in it. Setting up a smart home today involves a lot of manual customizations to make it function as the user wishes. What smart homes lack is the possibility to learn from users behaviour and habits in order to provide a customized environment for the user autonomously. The purpose of this thesis is to examine whether environmental data can be collected and used in a small smart home system to learn about the users behaviour. To collect data and attempt this learning process, a system is set up. The system uses a central control unit for mediation between wireless electrical outlets and sensors. The sensors track motion, light, temperature as well as humidity. The devices and sensors along with user interactions in the environment make up the collected data. Through studying the collected data, the system is able to create rules. These rules are used for the system to make decisions within its environment to suit the users’ needs. The performance of the system varies depending on how the data collection is handled. Results find that collecting data in intervals as well as when an action is made from the user is important. / Smarta hem är system avsedda för att hjälpa användare styra sin hemmiljö. Ett smart hem är uppbyggt av enheter med möjlighet att kommunicera med varandra. För att kontrollera enheterna i ett smart hem, används en central styrenhet. Att få ett smart hem att vara anpassat till användare är ansträngande och tidskrävande. Smarta hemsystem saknar i stor utsträckning möjligheten att lära sig av användarens beteende. Vad ett sådant lärande skulle kunna möjliggöra är ett skräddarsytt system utan användarens involvering. Syftet med denna avhandling är att undersöka hur användardata från en hemmiljö kan användas i ett smart hemsystem för att lära sig av användarens beteende. Ett litet smart hemsystem har skapats för att studera ifall denna inlärningsmetod är applicerbar. Systemet består av sensorer, trådlösa eluttag och en central styrenhet. Den centrala styrenheten används för att kontrollera de olika enheterna i miljön. Sensordata som sparas av systemet består av rörelse, ljusstyrka, temperatur och luftfuktighet. Systemet sparar även användarens beteende i miljön. Systemet skapar regler utifrån sparad data med målet att kunna styra enheterna i miljön på ett sätt som passar användaren. Systemets agerande varierade beroende på hur data samlades in. Resultatet visar vikten av att samla in data både i intervaller och när användare tar ett beslut i miljön.

Les pompiers méditerranéens sont confrontés à des embrasements soudains de la végétation (AFF) dont les mécanismes ne sont pas encore bien compris. La végétation étant l'unique combustible, nous nous sommes penchés sur les gaz qui en proviennent. Nous avons d’abord étudié la dégradation thermique de quatre Composés Organiques Volatils biogéniques (COVb) à l'aide d'une pyrolyse flash et d'un four tubulaire. À partir de cette étude et de la littérature, nous avons choisi un mélange d'étude afin expérimenter ses propriétés de combustion. Nous avons ainsi déterminé l'Énergie Minimale d’Inflammation (EMI) et la vitesse fondamentale de flamme de mélanges d'α-pinène/benzène qui sont respectivement les principaux COV détectés dans les plantes et dans les fumées de feux de forêt. Le dernier chapitre concerne l'étude stationnaire de l'accumulation de gaz dans des vallées à partir d'une maquette de forêt 1/400ème disposée dans une soufflerie. / Mediterranean firefighters cope with powerful accelerations of forest fires (AFF) whose mechanisms are not very well understood. Vegetation is the only fuel of forest fire, then we studied the gases coming from them. First, we studied the thermal degradation of four Biogenic Volatil Organic Compounds (BVOCs) thanks to a flash pyrolysis and a tubular oven. From this study and literature, we chose a representative VOC mixture to study its combustion properties. We determined Minimal Ignition Energy (MIE) and its laminar burning speed of mixtures of α-pinene/benzene that are respectively the main VOC detected in vegetation and forest fire smoke. The last chapter experiment the steady-state gas accumulation above a 1/400 V-shaped forest model.

A mobile laser scanning (MLS) system equipped with a lidar, inertial navigation system and satellite positioning can be used to reconstruct georeferenced point-based models of the surveyed environments. Ideal reconstruction requires accurate trajectories that are challenging to obtain in forests. Satellite signals are heavily degraded under the forest canopy, while lidar-based positioning is often inefficient due to the forest’s unstructured and complex nature. Most forestry-related solutions compute or improve the trajectory in post-processing, focusing on accuracy rather than the possibility of real-time operation. On the other hand, real-time solutions exist, but they are primarily tested and evaluated in urban environments, and the forest’s effect on them is less known. In this study, high-quality, real-time point-based forest model generation was considered by applying techniques from the field of robotics. Forest data were collected with an MLS system mounted 1) on a stick carried by a person and 2) mounted on a forest harvester while performing thinning operations. The system’s trajectory was computed using lidar-inertial-based smoothing and mapping algorithms with real-time limitations. In addition, satellite measurements were either fused into the smoothing algorithm contributing to the trajectory estimation or were used to georeference the trajectory in a post-processing manner. Collecting reliable reference trajectories is difficult in forests. Therefore, this study mainly contains qualitative and relative evaluation. The results indicate that real-time and onboard processing is feasible for forest data with adequate accuracy. State-of-the-art edge and planar feature-based lidar odometry was the most accurate but could not fully maintain real-time operation. On the other hand, the normal distributions transform-based odometry can maintain fast and constant computation with slightly lower accuracy. Fusing the satellite positioning for the mapping reduced the internal integrity of the reconstructed point cloud models, and it is suggested to use it for post-processed georeferencing instead. / Ett mobilt laserskanningssystem (MLS) utrustat med ett lidar, tröghetsnavigeringssystem och satellitpositionering kan användas för att rekonstruera georefererade punktbaserade modeller av de undersökta miljöerna. Idealisk återuppbyggnad kräver exakta banor som är utmanande att uppnå i skogar. Satellitsignaler är kraftigt försämrade under skogens tak, medan lidarbaserad positionering ofta är ineffektiv på grund av skogens ostrukturerade och komplexa natur. De flesta skogsbruksrelaterade lösningar beräknar eller förbättrar banan i efterbearbetning, med fokus på noggrannhet snarare än möjligheten till drift i realtid. Å andra sidan finns realtidslösningar, men de är främst testade och utvärderade i stadsmiljöer och skogens påverkan på dem är mindre känd. I denna studie övervägdes högkvalitativ, punktbaserad skogsmodellgenerering i realtid genom att tillämpa tekniker från robotteknikområdet. Skogsdata samlades in med ett MLS-system monterat 1) på en pinne som bärs av en person och 2) monterad på en skogsskördare under gallringsoperationer. Systemets bana beräknades med hjälp av lidar-tröghetsbaserade utjämnings- och kartläggningsalgoritmer med realtidsbegränsningar. Dessutom fuserades satellitmätningar antingen in i utjämningsalgoritmen som bidrog till banuppskattningen eller användes för att georeferera banan på ett efterbearbetningssätt. Att samla pålitliga referensbanor är svårt i skogar. Därför innehåller denna studie främst kvalitativ och relativ utvärdering. Resultaten indikerar att bearbetning i realtid och ombord är möjlig för skogsdata med tillräcklig noggrannhet. Toppmodern kant- och planfunktionsbaserad lidarodometri var den mest exakta men kunde inte helt upprätthålla realtidsdrift. Å andra sidan kan normalfördelningstransformationsbaserad odometri upprätthålla snabb och konstant beräkning med något lägre noggrannhet. Att sammansmälta satellitpositioneringen för kartläggningen minskade den interna integriteten hos de rekonstruerade punktmolnmodellerna, och det föreslås att man istället använder den för efterbehandlad georeferens.