One of the main challenges of the integration of unmanned aerial vehicles (UAVs) into today’s society is the risk of in-flight failures, such as motor failure, occurring in populated areas that can result in catastrophic accidents. We propose a framework to manage the consequences of an in-flight system failure and to bring down the aircraft safely without causing any serious accident to people, property, and the UAV itself. This can be done in three steps: a) Detecting a failure, b) Finding a safe landing spot, and c) Navigating the UAV to the safe landing spot. In this thesis, we will look at part b. Specifically, we are working to develop an active system that can detect landing sites autonomously without any reliance on UAV resources. To detect a safe landing site, we are using a deep learning algorithm named "You Only Look Once" (YOLO) that runs on a Jetson Xavier NX computing module, which is connected to a camera, for image processing. YOLO is trained using the DOTA dataset and we show that it can detect landing spots and obstacles effectively. Then by avoiding the detected objects, we find a safe landing spot. The effectiveness of this algorithm will be shown first by comprehensive simulations. We also plan to experimentally validate this algorithm by flying a UAV and capturing ground images, and then applying the algorithm in real-time to see if it can effectively detect acceptable landing spots.

In today's society, an enormous amount of data is created that is stored in various databases. Since the data is in many cases stored in different databases, there is a demand from organizations with a lot of data to be able to merge separated data and get an extraction of this resource. Extract, Transform and Load System (ETL) is a solution that has made it possible to easily merge different databases. However, the ETL market has been owned by large actors such as Amazon and Microsoft and the solutions offered are completely owned by these actors. This leaves the consumer with little ownership of the solution. Therefore, this thesis proposes a framework to create a component based ETL which gives consumers an opportunity to own and develop their own ETL solution that they can customize to their own needs. The result of the thesis is a prototype ETL solution that is built with the idea of being able to configure and customize the prototype and it accomplishes this by being independent of inflexible external libraries and a level of modularity that makes adding and removing components easy. The results of this thesis are verified with a test that shows how two different files containing data can be combined. / I dagens samhälle skapas det en enorm mängd data som är lagrad i olika databaser. Eftersom data i många fall är lagrat i olika databaser, finns det en efterfrågan från organisationer med mycket data att kunna slå ihop separerad data och få en utvinning av denna resurs. Extract, Transform and Load System (ETL) är en lösning som gjort det möjligt att slå ihop olika databaser. Dock är problemet denna expansion av ETL teknologi. ETL marknaden blivit ägd av stora aktörer såsom Amazon och Microsoft och de lösningar som erbjuds är helt ägda av dem. Detta lämnar konsumenten med lite ägodel av lösningen. Därför föreslår detta examensarbete ett ramverk för att skapa ett komponentbaserat ETL verktyg som ger konsumenter en möjlighet att utveckla en egen ETL lösning som de kan skräddarsy efter deras egna förfogande. Resultatet av examensarbete är en prototyp ETL-lösning som är byggd för att kunna konfigurera och skräddarsy prototypen. Lösningen lyckas med detta genom att vara oberoende av oflexibla externa bibliotek och en nivå av modularitet som gör addering och borttagning av komponenter enkelt. Resultatet av detta examensarbete är verifierat av ett test som visar på hur två olika filer med innehållande data kan kombineras.

Biological research has benefitted greatly from the advent of omic methods. For many biomolecules, mass spectrometry (MS) methods are most widely employed due to the sensitivity which allows low quantities of sample and the speed which allows analysis of complex samples. Improvements in instrument and sample preparation techniques create opportunities for large scale experimentation. The complexity and volume of data produced by modern MS-omic instrumentation challenges biological interpretation, while the complexity of the instrumentation, sample noise, and complexity of data analysis present difficulties in maintaining and ensuring data quality, validity, and relevance. We present a corpus of tools which improves quality assurance capabilities of instruments, provides comparison abilities for evaluating data analysis tool performance, distills ideas pertinent in MS analysis into a consistent nomenclature, enhances all lipid analysis by automatic structural classification, implements a rigorous and chemically derived lipid fragmentation prediction tool, introduces custom structural analysis approaches and validation techniques, simplifies protein analysis form SDS-PAGE sample excisions, and implements a robust peak detection algorithm. These contributions provide improved identification of biomolecules, improved quantitation, and improve data quality and algorithm clarity to the MS-omic field.

Increasing numbers of mathematics teachers must be recruited in coming years, because of a growing student population, teacher attrition, calls for smaller class size, and the need to replace out-of-subject teachers. Recruitment can be made more effective and efficient, if better information on career paths is provided to decision makers. This study attempts to analyze the academic decisions which lead to the outcome "becoming a mathematics teacher". Four groups were compared and contrasted: mathematics teachers, science teachers, other teachers, and non-teachers. Science teachers were removed from the "other teachers" category because of their many similarities to mathematics teachers on the variables examined. The question of whether these groups differ in ways that could help predict the outcome of interest was examined using the NCES dataset Baccalaureate &Beyond:93/97, which provides thousands of variables on academic path, demographics, and labor market histories for over 8,000 individuals. It was analyzed using the NCES online analytic tool DAS to generate tables showing percentage distribution of the four groups on variables organized according to the concepts demographics, family environment, academic path, and academic achievement. Further examination was conducted by entering the variables into a discriminant analysis. Mathematics teachers were found to differ from teachers of other K-12 fields on all of the four conceptual categories. However, only a few such differences were statistically significant. More significant differences were observed when the analyses were conducted separately for women and men. The trend observed was that those who became mathematics teachers were more likely to have attended public high schools and to have first attended two-year colleges; to have lower GPAs, more mathematics credits, and midrange CEE scores; and to be female.

Depth detection is a very common computer vision problem. It shows up primarily in robotics, automation, or 3D visualization domains, as it is essential for converting images to point clouds. One of the poster child applications is self driving cars. Currently, the best methods for depth detection are either very expensive, like LIDAR, or require precise calibration, like stereo cameras. These costs have given rise to attempts to detect depth from a monocular camera (a single camera). While this is possible, it is harder than LIDAR or stereo methods since depth can't be measured from monocular images, it has to be inferred. A good example is covering one eye: you still have some idea how far away things are, but it's not exact. Neural networks are a natural fit for this. Here, we build on previous neural network methods by applying a recent state of the art model to panoramic images in addition to pinhole ones and performing a comparative evaluation. First, we create a simulated depth detection dataset that lends itself to panoramic comparisons and contains pre-made cylindrical and spherical panoramas. We then modify monodepth2 to support cylindrical and cubemap panoramas, incorporating current best practices for depth detection on those panorama types, and evaluate its performance for each type of image using our dataset. We also consider the resources used in training and other qualitative factors.

Churn prediction is a common data mining problem that many companies face across industries. More commonly, customer churn has been studied extensively within the telecommunications industry where there is low customer retention due to high market competition. Similar to customer churn, employee churn is very costly to a company and by not deploying proper risk mitigation strategies, profits cannot be maximized, and valuable employees may leave the company. The cost to replace an employee is exponentially higher than finding a replacement, so it is in any company’s best interest to prioritize employee retention. This research combines machine learning techniques with undersampling in hopes of identifying employees at risk of churn so retention strategies can be implemented before it is too late. Four different classification algorithms are tested on a variety of undersampled datasets in order to find the most effective undersampling and classification method for predicting employee churn. Statistical analysis is conducted on the appropriate evaluation metrics to find the most significant methods. The results of this study can be used by the company to target individuals at risk of churn so that risk mitigation strategies can be effective in retaining the valuable employees. Methods and results can be tested and applied across different industries and companies.

Breast cancer is the most common type of cancer in the world, and it is the second deadliest cancer for females. In the fight against breast cancer, early detection plays a large role in saving people’s lives. In this work, an image classifier is designed to diagnose breast tumors as benign or malignant. The classifier is designed with a neural network and trained on the BreakHis dataset. After creating the initial design, a variety of methods are used to try to improve the performance of the classifier. These methods include preprocessing, increasing the number of training epochs, changing network architecture, and data augmentation. Preprocessing includes changing image resolution and trying grayscale images rather than RGB. The tested network architectures include VGG16, ResNet50, and a custom structure. The final algorithm creates 50 classifier models and keeps the best one. Classifier designs are primarily judged on the classification accuracies of their best model and their median model. Designs are also judged on how consistently they produce their highest performing models. The final classifier design has a median accuracy of 93.62% and best accuracy of 96.35%. Of the 50 models generated, 46 of them performed with over 85% accuracy. The final classifier design is compared to the works of two groups of researchers who created similar classifiers for the same dataset. This will show that the classifier performs at the same level or better than the classifiers designed by other researchers. The classifier achieves similar performance to the classifier made by the first group of researchers and performs better than the classifier from the second. Finally, the learned lessons and future steps are discussed.

Radio Frequency Machine Learning (RFML) is the application of ML techniques to solve problems in the RF domain as an alternative to traditional digital-signal processing (DSP) techniques. Notable among these are the tasks of specific emitter identification (SEI), determining source identity of a received RF signal, and automated modulation classification (AMC), determining the modulation scheme of a received RF transmission. Both tasks have a number of algorithms that are effective on simulated data, but struggle to generalize to data collected in the real-world, partially due to the lack of available datasets upon which to train models and understand their limitations. This thesis covers the practical considerations for systems that can create high-quality datasets for RFML tasks, how variances from real-world effects in these datasets affect RFML algorithm performance, and how well models developed from these datasets are able to generalize and adapt across different receiver hardware platforms. Moreover, this thesis presents a proof-of-concept system for large-scale and efficient data generation, proven through the design and implementation of a custom platform capable of coordinating transmissions from nearly a hundred Software-Defined Radios (SDRs). This platform was used to rapidly perform experiments in both RFML performance sensitivity analysis and successful transfer between SDRs of trained models for both SEI and AMC algorithms. / Master of Science / Radio Frequency Machine Learning (RFML) is the application of machine learning techniques to solve problems having to do with radio signals as an alternative to traditional signal processing techniques. Notable among these are the tasks of specific emitter identification (SEI), determining source identity of a received signal, and automated modulation classification (AMC), determining the data encoding format of a received RF transmission. Both tasks have practical limitations related to the real-world collection of RF training data. This thesis presents a proof-of-concept for large-scale, efficient data generation and management, as proven through the design and construction of a custom platform capable of coordinating transmissions from nearly a hundred radios. This platform was used to rapidly perform experiments in both RFML performance sensitivity analysis and successful cross-radio transfer of trained behaviors.

Demand forecasting for sales is a widely researched topic that is essential for a business to prepare for market changes and increase profits. Existing research primarily focus on data that is more suitable for machine learning applications compared to the data accessible to companies lacking prior machine learning experience. This thesis performs demand forecasting on a known sales dataset and a dataset accessed directly from such a company, in the hopes of gaining insights that can help similar companies better utilize machine learning in their business model. LigthGBM, Linear Regression and Random Forest models are used along with several regression error metrics and plots to compare the performance of the two datasets. Both data sets are preprocessed into the same structure based on equivalent features found in each set. The company dataset is determined to be unfit for machine learning forecasting even after preprocessing measures and multiple possible reasons are established. The main contributors are a lack of observations per article and uniformity through the time series.

Automatic machine learning (AutoML) systems have been shown to perform better when they use metamodels trained offline. Existing offline metalearning approaches treat ML models as black boxes. However, modern ML models often compose multiple ML algorithms into ML pipelines. We expand previous metalearning work on estimating the performance and ranking of ML models by exploiting the metadata about which ML algorithms are used in a given pipeline. We propose a dynamically assembled neural network with the potential to model arbitrary DAG structures. We compare our proposed metamodel against reasonable baselines that exploit varying amounts of pipeline metadata, including metamodels used in existing AutoML systems. We observe that metamodels that fully exploit pipeline metadata are better estimators of pipeline performance. We also find that ranking pipelines based on dataset metafeature similarity outperforms ranking based on performance estimates.