Semantic segmentation of a scene aims to give meaning to the scene by dividing it into meaningful — semantic — parts. Understanding the scene is of great interest for all kinds of autonomous systems, but manual annotation is simply too time consuming, which is why there is a need for an alternative approach. This thesis investigates the possibility of automatically segmenting 3D-models of urban scenes, such as buildings, into a predetermined set of labels. The approach was to first acquire ground truth data by manually annotating five 3D-models of different urban scenes. The next step was to extract features from the 3D-models and evaluate which ones constitutes a suitable feature space. Finally, three supervised learners were implemented and evaluated: k-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Random Classification Forest (RCF). The classifications were done point-wise, classifying each 3D-point in the dense point cloud belonging to the model being classified. The result showed that the best suitable feature space is not necessarily the one containing all features. The KNN classifier got the highest average accuracy overall models — classifying 42.5% of the 3D points correct. The RCF classifier managed to classify 66.7% points correct in one of the models, but had worse performance for the rest of the models and thus resulting in a lower average accuracy compared to KNN. In general, KNN, SVM, and RCF seemed to have different benefits and drawbacks. KNN is simple and intuitive but by far the slowest classifier when dealing with a large set of training data. SVM and RCF are both fast but difficult to tune as there are more parameters to adjust. Whether the reason for obtaining the relatively low highest accuracy was due to the lack of ground truth training data, unbalanced validation models, or the capacity of the learners, was never investigated due to a limited time span. However, this ought to be investigated in future studies.

Goal of this work is to make acquaintance and study meta-learningu methods, program algorithm and compare with other machine learning methods.

Thesis focuses on disk arrays, where the goal is to design test scenarios to measure performance of disk array and use predictive analytics tools to train a model that will predict the selected performance parameter on a measured set of data. The implemented web application demonstrates the functionality of the trained model and shows estimate of the disk array performance.

The ability to reliably connect electrical cables is important in many applications. A poor connection can become a fire hazard, so it is important that cables are always appropriately connected. This thesis investigates methods for monitoring of a machine that presses cable connectors onto cables. Using sensor data from the machine, would it be possible to create an algorithm that can automatically identify the cable and connector and thus make decisions on how a connector should be pressed for successful attachment? Furthermore, would it be possible to create an anomaly detection algorithm that is able to detect whether a connector has been incorrectly pressed by the end user? If these two questions can be addressed, the solutions would minimise the likelihood of errors, and enable detection of errors that anyway do arise. In this thesis, it is shown that the k-Nearest Neighbour (kNN) algorithm and Long Short-Term Memory (LSTM) network are both successful in classification of connectors and cables, both performing with 100% accuracy on the test set. The LSTM is the more promising alternative in terms of convergence and speed, being 28 times faster as well as requiring less memory. The distance-based methods and an autoencoder are investigated for the anomaly detection task. Data corresponding to a wide variety of possible incorrect kinds of usage of the tool were collected. The best anomaly detector detects 92% of incorrect cases of varying degrees of difficulty, a number which was higher than expected. On the tasks investigated, the performance of the neural networks are equal to or higher than the performance of the alternative methods.

<p>Emergency services (ES) pilots operate in a dynamic, high-risk team environment, as a subset of general aviation (GA) operations. The time constraints associated with ES operations means that ES pilots must make flight decisions quickly and often with limited or incomplete information (Worm, 1999). Due to the nature of ES operations, the consequences of an incorrect flight decision can be severe, including loss of life. ES operations are often initiated by extreme weather events, and ES pilots are frequently required to fly on the boundary between marginal visual flight rules (MVFR) weather conditions and instrument meteorological conditions (IMC). Unfortunately, an unintended transition into IMC is the leading cause of fatal accidents in GA operations (Ayiei et al., 2020). Mission objectives dictate that most ES pilots fly below 1,500’ above ground level (AGL) for extended periods of time, and low-altitude flight in hazardous weather can reduce a pilot’s outside visual reference, thus leading to spatial disorientation, loss of control, or controlled flight into terrain. To mitigate this problem, ES pilots must be able to accurately assess weather conditions before and during flight. However, the current method of presenting meteorological aerodrome reports (METARs) on weather displays can be misleading to pilots. Weather conditions in the areas between weather observation stations can be different than what is reported by the METAR observations at those stations. This can cause current or forecasted weather conditions <em>between</em> weather stations to be incompletely represented. However, pilots are given no obvious indication of how incompletely represented weather conditions can affect weather-related risk. This research demonstrates that a <em>Kth</em> Nearest Neighbor (KNN) analysis can be used to identify areas where the variability of conditions between weather stations (and thus weather-related risk) is incompletely represented by METAR observations. In addition, it is shown that areas where there is an increased risk of an unintended transition from MVFR to IMC can be identified among areas with incompletely represented conditions and depicted to pilots on aviation weather displays. Machine learning tactics are proposed as a way to consider additional inputs in future KNN analyses, and several emerging technologies are proposed as mediums to collect additional weather observations. The ability for an ES pilot to more accurately assess weather-related risk in MVFR conditions using the proposed technologies is evaluated, the benefits to ES pilots and the GA community are discussed, and the requirements and limitations of the study are examined.</p>

Den stora mängden tillgänglig data på internet kan användas för att förbättra förutsägelser genom maskininlärning. Problemet är att sådan data ofta är i ett obehandlat format och kräver att någon manuellt bestämmer etiketter på den insamlade datan innan den kan användas av algoritmen. Semi-supervised learning (SSL) är en teknik där algoritmen använder ett fåtal förbehandlade exempel och därefter automatiskt bestämmer etiketter för resterande data. Ett tillvägagångssätt inom SSL är att representera datan i en graf, vilket kallas för graf-baserad semi-supervised learning (GSSL), och sedan hitta likheter mellan noderna i grafen för att automatiskt bestämma etiketter.Vårt mål i denna uppsatsen är att förenkla de avancerade processerna och stegen för att implementera en GSSL-algoritm. Vi kommer att gå igen grundläggande steg som hur utvecklingsmiljön ska installeras men även mer avancerade steg som data pre-processering och feature extraction. Feature extraction metoderna som uppsatsen använder sig av är bag-of-words (BOW) och term frequency-inverse document frequency (TF-IDF). Slutgiltligen presenterar vi klassificering av dokument med Label Propagation (LP) och Multinomial Naive Bayes (MNB) samt en detaljerad beskrivning över hur GSSL fungerar.Vi presenterar även prestanda för klassificering-algoritmerna genom att klassificera 20 Newsgroup datasetet med LP och MNB. Resultaten dokumenteras genom två olika utvärderingspoäng vilka är F1-score och accuracy. Vi gör även en jämförelse mellan MNB och LP med två olika typer av kärnor, KNN och RBF, på olika mängder av förbehandlade träningsdokument. Resultaten ifrån klassificering-algoritmerna visar att MNB är bättre på att klassificera datasetet än LP. / The large amount of available data on the web can be used to improve the predictions made by machine learning algorithms. The problem is that such data is often in a raw format and needs to be manually labeled by a human before it can be used by a machine learning algorithm. Semi-supervised learning (SSL) is a technique where the algorithm uses a few prepared samples to automatically prepare the rest of the data. One approach to SSL is to represent the data in a graph, also called graph-based semi-supervised learning (GSSL), and find similarities between the nodes for automatic labeling.Our goal in this thesis is to simplify the advanced processes and steps to implement a GSSL-algorithm. We will cover basic tasks such as setup of the developing environment and more advanced steps such as data preprocessing and feature extraction. The feature extraction techniques covered are bag-of-words (BOW) and term frequency-inverse document frequency (TF-IDF). Lastly, we present how to classify documents using Label Propagation (LP) and Multinomial Naive Bayes (MNB) with a detailed explanation of the inner workings of GSSL. We showcased the classification performance by classifying documents from the 20 Newsgroup dataset using LP and MNB. The results are documented using two different evaluation scores called F1-score and accuracy. A comparison between MNB and the LP-algorithm using two different types of kernels, KNN and RBF, was made on different amount of labeled documents. The results from the classification algorithms shows that MNB is better at classifying the data than LP.

With an overwhelming number of movies available globally, it can be a daunting task for users to find movies that cater to their individual preferences. The vast selection can often leave people feeling overwhelmed, making it challenging to pick a suitable movie. As a result, movie service providers need to offer a recommendation system that adds value to their customers. A movie recommendation system can help customers in this regard by providing a process that assists in finding movies that match their preferences. Previous studies on recommendation systems that use Machine Learning (ML) algorithms have demonstrated that these algorithms outperform some of the existing recommendation methods regarding recommendation strategy. However, there is still room for further improvement, especially when it comes to exploring scenarios where users need to spend a considerable amount of time finding movies related to their preferred genres. This prolonged search for the right movies can give rise to problems such as data sparsity and cold start. To address these issues, we propose a machine learning-based recommender system for movie genres using the K-nearest Neighbours (KNN) algorithm. Our final system utilizes a slider bar on a Streamlit web app, allowing users to select their preferred movies and see recommendations for similar movies. By incorporating user preferences, our system provides personalized recommendations that are more likely to meet the user's interests and preferences. To address our research question: “How and to what extent can a machine learning-based recommender system be developed focusing on movie genres where movie popularity can be predicted based on its content?” we propose three main research objectives. Firstly, we investigate the employment of a classification algorithm in recommending movies focusing on interest genres. Secondly, we evaluate the performance of our classification algorithm concerning movie viewers. Thirdly, we represent the popularity of movie genres based on the content and investigate how this representation can inform the movie recommendation algorithm. On the heels of an experimental strategy, we extract and pre-process a dataset of movies and their associated genre labels from Kaggle. The dataset consists of two files derived from The Movie Database (TMDB) 5000 Movie Dataset. We develop a machine learning-based recommender system based on the similarity of movie genres using the extracted and pre-processed dataset. We vary the KNN algorithm with a slider bar to recommend movies of varying similarity to the selected movie, ranging from similar to diverse in genre. This approach can suggest movies with different titles for users with diverse preferences. We evaluate the performance of the KNN classification algorithm using a user's interest genres, measuring its accuracy, precision, recall, and F1-score. The algorithm's accuracy ranges from low to moderate across different values of K, indicating its moderate effectiveness in predicting user preferences. The algorithm's precision ranges from moderate to high, implying that it provides accurate recommendations to the user. The recall score improves with increasing K and reaches its maximum at K=15, demonstrating its ability to retrieve relevant recommendations. The algorithm achieves a good balance between precision and recall, with an average F1-score of 0.60. This means that the algorithm can accurately identify relevant movies and recommend them to users with a high degree of accuracy. Furthermore, our result shows that the popularity visualization technique using KNN is a powerful tool for analysing and understanding the popularity of different movie genres, which can inform important decisions related to marketing, distribution, and production in the movie industry. In conclusion, our machine learning-based recommender system using KNN for movie genres is a game changer. It allows users to select their preferred movies and see recommendations for similar movies using a slider bar on a Streamlit web app. If confirmed by future research, the promising findings of this thesis can pave the way for developing and incorporating other classification algorithms and features for movie recommendation and evaluation. Furthermore, the adjustable slider bar ranges on the Streamlit web app allow users to customize their movie preferences and receive tailored recommendations.

With new 2,4-Dichlorophenoxyacetic acid (2,4-D) tolerant crops, increases in off-target movement events are expected. New formulations may mitigate these events, but standard lab techniques are ineffective in identifying these 2,4-D formulations. Using Fourier-transform infrared spectroscopy and machine learning algorithms, research was conducted to classify 2,4-D formulations in treated herbicide-tolerant soybeans and cotton and observe the influence of leaf treatment status and collection timing on classification accuracy. Pooled Classification models using k-nearest neighbor classified 2,4-D formulations with over 65% accuracy in cotton and soybean. Tissue collected 14 DAT and 21 DAT for cotton and soybean respectively produced higher accuracies than the pooled model. Tissue directly treated with 2,4-D also performed better than the pooled model. Lastly, models using timing and treatment status as factors resulted in higher accuracies, with cotton 14 DAT New Growth and Treated models and 28 DAT and 21 DAT Treated soybean models achieving the best accuracies.

Falls are a major cause of morbidity and mortality in neurological disorders. Technical means of detecting falls are of high interest as they enable rapid notification of caregivers and emergency services. Such approaches must reliably differentiate between normal daily activities and fall events. A promising technique might be based on the classification of movements based on accelerometer signals by machine-learning algorithms, but the generalizability of classifiers trained on laboratory data to real-world datasets is a common issue. Here, three machine-learning algorithms including Support Vector Machine (SVM), k-Nearest Neighbors (kNN), and Random Forest (RF) were trained to detect fall events. We used a dataset containing intentional falls (SisFall) to train the classifier and validated the approach on a different dataset which included real-world accidental fall events of elderly people (FARSEEING). The results suggested that the linear SVM was the most suitable classifier in this cross-dataset validation approach and reliably distinguished a fall event from normal everyday activity at an accuracy of 93% and similarly high sensitivity and specificity. Thus, classifiers based on linear SVM might be useful for automatic fall detection in real-world applications.

Made available in DSpace on 2014-07-29T14:57:46Z (GMT). No. of bitstreams: 1 dissertacao-fernando.pdf: 677510 bytes, checksum: 19704f0b04ee313a63b053f7f9df409c (MD5) Previous issue date: 2010-10-10 / Most research on Automatic Text Categorization (ATC) seeks to improve the classifier performance (effective or efficient) responsible for automatically classifying a document d not yet rated. The k nearest neighbors (kNN) is simpler and it s one of automatic classification methods more effective as proposed. In this paper we proposed two kNN variations, Inverse kNN (kINN) and Symmetric kNN (kSNN) with the aim of improving the effectiveness of ACT. The kNN, kINN and kSNN methods were applied in Reuters, 20ng and Ohsumed collections and the results showed that kINN and kSNN methods were more effective than kNN method in Reuters and Ohsumed collections. kINN and kSNN methods were as effective as kNN method in 20NG collection. In addition, the performance achieved by kNN method is more stable than kINN and kSNN methods when the value k change. A parallel study was conducted to generate new features in documents from the similarity matrices resulting from the selection criteria for the best results obtained in kNN, kINN and kSNN methods. The SVM (considered a state of the art method) was applied in Reuters, 20NG and Ohsumed collections - before and after applying this approach to generate features in these documents and the results showed statistically significant gains for the original collection. / Grande parte das pesquisas relacionadas com a classificação automática de textos (CAT) tem procurado melhorar o desempenho (eficácia ou eficiência) do classificador responsável por classificar automaticamente um documento d, ainda não classificado. O método dos k vizinhos mais próximos (kNN, do inglês k nearest neighbors) é um dos métodos de classificação automática mais simples e eficazes já propostos. Neste trabalho foram propostas duas variações do método kNN, o kNN invertido (kINN) e o kNN simétrico (kSNN) com o objetivo de melhorar a eficácia da CAT. Os métodos kNN, kINN e kSNN foram aplicados nas coleções Reuters, 20NG e Ohsumed e os resultados obtidos demonstraram que os métodos kINN e kSNN tiveram eficácia superior ao método kNN ao serem aplicados nas coleções Reuters e Ohsumed e eficácia equivalente ao método kNN ao serem aplicados na coleção 20NG. Além disso, nessas coleções foi possível verificar que o desempenho obtido pelo método kNN é mais estável a variação do valor k do que os desempenhos obtidos pelos métodos kINN e kSNN. Um estudo paralelo foi realizado para gerar novas características em documentos a partir das matrizes de similaridade resultantes dos critérios de seleção dos melhores resultados obtidos na avaliação dos métodos kNN, kINN e kSNN. O método SVM, considerado um método de classificação do estado da arte em relação à eficácia, foi aplicado nas coleções Reuters, 20NG e Ohsumed - antes e após aplicar a abordagem de geração de características nesses documentos e os resultados obtidos demonstraram ganhos estatisticamente significativos em relação à coleção original.