This thesis aims to show that a Stacking Ensemble of multiple base-learners can provide a more accurate prediction of commercial flight delays between the ten largest US airports than the individual prediction models. Three types of machine learning models, namely LASSO, Random Forests and Neural Networks are used as base-learners with different hyper- parameters. A Stacking Ensemble is created by using LASSO as meta-learner. The Stacking Ensemble and the base-learners that performed best on the training data are then evaluated on a test data set. The results are compared by the metrics accuracy, ROC AUC, MCC and F1 Score. It is shown that the Stacking Ensemble is able to provide superior predictions for flight delays in comparison to the best individual models.

Predicting and characterizing the state of the ocean is needed for various scientific, industrial, social, management, and recreational activities. Despite the tremendous progress in ocean modeling and simulation capabilities, the ocean models still suffer from different sources of uncertainties. To obtain accurate ocean state predictions, data assimilation (DA) is widely used to constrain the ocean model outputs with available observations. Ensemble Kalman filtering (EnKF) is a sequential DA approach that represents the distribution of the system state through an ensemble of ocean state samples. Different factors may limit the performance of an EnKF in realistic ocean applications, particularly the use of small ensembles and poorly known model error statistics, and also to a lesser extent the strongly nonlinear variations and abrupt regime changes, and unsatisfied underlying assumptions such as the commonly used white observation noise assumption. The objective of this PhD thesis is to develop, implement and test efficient ensemble filtering schemes to enhance the performances of EnKFs in such challenging settings. We resort to the one-step-ahead (OSA) smoothing formulation of the Bayesian filtering problem to introduce EnKFs involving a new update step with future observations (smoothing) between two successive analyses, thereby conditioning the ensemble sampling with more information. We show that this approach enhances the EnKFs performances by providing improved ensemble background statistics, and showcase its performance with realistic ocean DA and forecasting applications, namely a storm surge EnKF forecasting system and the Red Sea ensemble DA and forecasting system. We then derive new EnKF-based schemes accounting for time-correlated observation errors for efficient DA into the class of large dimensional DA problems where observation errors statistics are correlated in time, and further propose a new approach for online estimation of the parameters of the observation error time-correlations model concurrently with the state. We also exploit the OSA-smoothing formulation to propose a new joint EnKF with OSA-smoothing which mitigates for the reported inconsistencies in the joint EnKF update for efficient DA into one-way-coupled systems.

This thesis addresses the problem of Road Damage Detection using object detection models,Yolov8 and Yolov5. While Yolov5 has been utilized in prior road damage detection projects, thiswork introduces the application of the newly released Yolov8 model to this domain. We haveprepared a dataset of 3,000 annotated images of road damage in Sweden and applied variousYolov8 and Yolov5 models to this dataset and a larger international one. The potential ofdeploying a lightweight Yolov8 model in a smartphone application for real-time detection, aswell as the effectiveness of an ensemble approach combining several models, were alsoexplored. The results show an F1 score of 0.57 and 0.6 for the best-performing models on theSwedish dataset and an international Road damage dataset respectively. Several box clusteringmethods were tested to combine the predictions of the ensemble, but none outperformed thebest individual model. A Quantized version of Yolov8 was deployed to a smartphone device withsatisfying performance. This work aims to create a model which can ultimately be used toimprove road safety and quality.T

Two very different classes of theoretical models have been proposed to explain visual perception. One class of models assume that there is a point at which we become consciously aware of a stimulus, known as a threshold. This threshold is the foundation of discrete process models all of which describe an all-or-none transition between the mental state of perceiving a stimulus and the state of not perceiving a stimulus. In contrast, the other class of models assume that mental states change continuously. These continuous models are founded in signal detection theory and the more contemporary models in Bayesian inference frameworks. The continuous model is the more widely accepted model of perception, and as such discrete process models were mostly discarded. Nonetheless, there has been a renewed debate on continuous versus discrete perception, and recent work has renewed the idea that perception can be all-or-none. In this dissertation, we developed an experimental platform and modeling framework to test whether visual perception exhibits measurable characteristics consistent with discrete perception. The results of this study revealed a selective influence of stimulus type on the way that a visual stimulus is processed. Moreover this selective influence implied perception can either be discrete or continuous depending on the underlying perceptual processing. These qualitative differences in the way perception occurs even for highly similar stimuli such as motion or orientation have crucial implications for models of perception, as well as our understanding of neurophysiology and conscious perception.

ABSTRACT MAJID MEMARI, for the Masters of Science degree in Computer Science, presented on November 3rd, 2017 at Southern Illinois University, Carbondale, IL. Title: PREDICTING THE STOCK MARKET USING NEWS SENTIMENT ANALYSIS Major Professor: Dr. Norman Carver Big data is a term for data sets that are so large or complex that traditional data processing application software is inadequate to deal with them. GDELT is the largest, most comprehensive, and highest resolution open database ever created. It is a platform that monitors the world's news media from nearly every corner of every country in print, broadcast, and web formats, in over 100 languages, every moment of every day that stretches all the way back to January 1st, 1979, and updates daily [1]. Stock market prediction is the act of trying to determine the future value of a company stock or other financial instrument traded on an exchange. The successful prediction of a stock's future price could yield significant profit. The efficient-market hypothesis suggests that stock prices reflect all currently available information and any price changes that are not based on newly revealed information thus are inherently unpredictable [2]. On the other hand, other studies show that it is predictable. The stock market prediction has been a long-time attractive topic and is extensively studied by researchers in different fields with numerous studies of the correlation between stock market fluctuations and different data sources derived from the historical data of world major stock indices or external information from social media and news [6]. The main objective of this research is to investigate the accuracy of predicting the unseen prices of the Dow Jones Industrial Average using information derived from GDELT database. Dow Jones Industrial Average (DJIA) is a stock market index, and one of several indices created by Wall Street Journal editor and Dow Jones & Company co-founder Charles Dow. This research is based on data sets of events from GDELT database and daily prices of the DJI from Yahoo Finance, all from March 2015 to October 2017. First, multiple different classification machine learning models are applied to the generated datasets and then also applied to multiple different Ensemble methods. In statistics and machine learning, Ensemble methods use multiple learning algorithms to obtain better predictive performance than could be obtained from any of the constituent learning algorithms alone. Afterwards, performances are evaluated for each model using the optimized parameters. Finally, experimental results show that using Ensemble methods has a significant (positive) impact on improving the prediction accuracy. Keywords: Big Data, GDELT, Stock Market, Prediction, Dow Jones Index, Machine Learning, Ensemble Methods

River ice breakup is a vital process to the morphology and hydrology of many rivers in Canada, often governing peak flows of the river. These events can occur through multiple mechanisms, with the potential for volatile or early breakup events that can have severe impacts to the river. Ice jam flooding can be a potentially devastating result of river ice breakup while early breakup of ice cover in a mid-winter breakup can be unpredictable and greatly alter the remaining ice season. These events are growing increasingly common as a result of climate change, and as a result there is a need to develop prediction tools for these events to aid in decision making support. Past investigations into developing such tools, especially from a data-driven modelling perspective, are challenged by the availability and complexity of the data related to these rare and dangerous to measure events. Therefore, the goal of this dissertation was to develop and apply methods to address the historical challenges and shortcomings in predicting these events through the use of data-driven modelling techniques. This includes: i) development of a stacking ensemble modelling framework for the prediction of ice jam presence during the spring breakup season of a river, utilising variable selection and rare-event forecasting techniques in combination with a comprehensive selection of machine-learning algorithms; ii) return period and trend analysis of mid-winter breakups in conjunction with comprehensive input analysis techniques to identify the key drivers of these events’ severity and develop a means of classifying the flood risk based on hydroclimatic traits; iii) the development of a two-level modelling system for the prediction of the occurrence and timing of mid-winter breakups on a national scale utilising rare event forecasting techniques and imbalanced learning; and iv) development of a novel hybrid semantic and machine learning modelling system in which an ontology is used in conjunction with network analysis techniques to select variables for machine learning models, which is used on a national case study of the prediction of spring breakup timing in Canada. The results of each study in application to their respective case studies demonstrate the effectiveness of the proposed techniques, which are shown to be easily adaptable to other regions or locations. These techniques can form the backbone of decision-making support for communities on rivers that are affected by the unpredictable and oftentimes volatile nature of river ice breakup. / Thesis / Candidate in Philosophy / River ice breakup is a key event to the hydrology of rivers throughout Canada, playing a major role in their physical and ecological characteristics. The timing and mechanism of these events can, however, be unpredictable and volatile, with the effects of climate change only exacerbating these risks. This dissertation focuses on addressing these potential issues through the application of machine learning and hybrid modeling in the prediction of river ice breakup events. Advanced data driven techniques coupled with novel applications of other analytical methods are used to: i) predict the presence of ice jams through the application of stacking ensemble modelling; ii) predict the severity of mid-winter breakups through application of trend and variable analysis; iii) predict the occurrence and timing of mid-winter breakups using rare-event forecasting techniques; and iv) develop a novel hybrid modelling scheme coupling ontology-based semantic modelling and machine learning to predict spring breakup timing. Detailed case studies for each application are provided demonstrating the effectiveness of the discussed techniques.

No description available.

No description available.

No description available.

No description available.