Exploring a Flooding-Sensors-Agnostic Prediction of the Damage Consequences Based on Machine Learning

Recently, progressive flooding simulations have been applied onboard to support decisions during emergencies based on the outcomes of flooding sensors. However, only a small part of the existing fleet of passenger ships is equipped with flooding sensors. In order to ease the installation of emergency decision support systems on older vessels, a flooding-sensor-agnostic solution is advisable to reduce retrofit cost. In this work, the machine learning algorithms trained with databases of progressive flooding simulations are employed to assess the main consequences of a damage scenario (final fate, flooded compartments, time-to-flood). Among the others, several classification techniques are here tested using as predictors only the time evolution of the ship floating position (heel, trim and sinkage). The proposed method has been applied to a box-shaped barge showing promising results. The promising results obtained applying the bagged decision trees and weighted k-nearest neighbours suggests that this new approach can be the base for a new generation of onboard decision support systems.

Integration of Deep and Ensemble Learning for Detecting COVID-19 in Computed Tomography Images

This paper presents an approach for detecting covid-19 in Computed Tomography (CT) images by integrating deep convolutional neural networks and ensembles of decision trees. The proposed approach consisted of three steps. In the first step, the CT images slices were collected and processed. In the second step, a deep convolutional neural network was trained to predict covid-19 in the CT images. In the third step, deep features were extracted and were used to train an ensemble of decision trees. Six types/packages of ensembles of decision trees were investigated: extreme gradient boosting (XGBoost), bagged decision trees (BDT), random forest (RF), adaptive boosting decision trees (Adaboost), gradient boosting decision trees (GBDT), and dropouts meet multiple additive regression trees (DART). The accuracy, sensitivity, specificity, f1-score, precision, and area under the ROC curve (AUC) were calculated to compare the models against each other. The proposed approach revealed the highest performance with a RF that reported 0.87 accuracy, 0.87 f1-score, and 0.90 AUC. The developed models revealed similar performance when compared to previously published models. This highlights the efficiency of combining deep networks with ensembles of decision trees for detecting covid-19.

Integration of Deep and Ensemble Learning for Detecting COVID-19 in Computed Tomography Images

This paper presents an approach for detecting covid-19 in Computed Tomography (CT) images by integrating deep convolutional neural networks and ensembles of decision trees. The proposed approach consisted of three steps. In the first step, the CT images slices were collected and processed. In the second step, a deep convolutional neural network was trained to predict covid-19 in the CT images. In the third step, deep features were extracted and were used to train an ensemble of decision trees. Six types/packages of ensembles of decision trees were investigated: extreme gradient boosting (XGBoost), bagged decision trees (BDT), random forest (RF), adaptive boosting decision trees (Adaboost), gradient boosting decision trees (GBDT), and dropouts meet multiple additive regression trees (DART). The accuracy, sensitivity, specificity, f1-score, precision, and area under the ROC curve (AUC) were calculated to compare the models against each other. The proposed approach revealed the highest performance with a RF that reported 0.87 accuracy, 0.87 f1-score, and 0.90 AUC. The developed models revealed similar performance when compared to previously published models. This highlights the efficiency of combining deep networks with ensembles of decision trees for detecting covid-19.

Contribution of clouds radiative forcing to the local surface temperature variability

<p>The local contribution of clouds to the surface energy balance and temperature variability is an important topic in order to apprehend how this intake affects local climate variability and extreme events, how this contribution varies from one place to another, and how it evolves in a warming climate. The scope of this study is to understand how clouds impact temperature variability, to quantify their contribution, and to compare their effects to other surface processes. To do so, we develop a method to estimate the different terms that control temperature variability at the surface (∂T<sub>2m</sub> /∂t) by using this equation: <strong>∂T<sub>2m</sub> /∂t=R+HA+HG+Adv</strong> where R is the radiation that is separated into the cloud term (R<sub>cloud</sub>) and the clear sky one (R<sub>CS</sub>), HA the atmospheric heat exchange, HG the ground heat exchange, and Adv the advection. These terms are estimated hourly, almost only using direct measurements from SIRTA-ReOBS dataset (an hourly long-term multi-variables dataset retrieved from SIRTA, an observatory located in a semi-urban area 20-km South-West of Paris; Chiriaco et al., 2019) for a five-years period. The method gives good results for the hourly temperature variability, with a 0.8 correlation coefficient and a weak residual term between left part (directly measured) and right part of the equation.</p><p>A bagged decision trees analysis of this equation shows that R<sub>CS</sub> dominates temperature variability during daytime and is mainly modulated by cloud radiative effect (R<sub>cloud</sub>). During nighttime, the bagged decision trees analysis determines that R<sub>cloud</sub> is the term controlling temperature changes. When a diurnal cycle analysis (split into seasons) is performed for each term, HA becomes an important negative modulator in the late afternoon, chiefly in spring and summer, when evaporation and thermal conduction are increased. In contrast, HG and Adv terms do not play an essential role on temperature variability at this temporal scale and their contribution is barely considerable in the one-hour variability, but still they remain necessary in order to obtain the best coefficient estimator between the directly measured observations and the method estimated. All terms except advection have a marked monthly-hourly cycle.</p><p>Next steps consist in characterize the types of clouds and study their physical properties corresponding to the cases where R<sub>cloud</sub> is significant, using the Lidar profiles also available in the SIRTA-ReOBS dataset.</p>

A Fourier-Bessel Expansion-Based Method for Automated Detection of Atrial Fibrillation From Electrocardiogram Signals

This chapter presents a new methodology for detection and identification of cardiovascular diseases from a single-lead electrocardiogram (ECG) signal of short duration. More specifically, this method deals with the detection of the most common cardiac arrhythmia called atrial fibrillation (AF) in noisy and non-clinical environment. The method begins with appropriate pre-processing of ECG signals in order to get the RR-interval and heart rate (HR) signals from them. A set of indirect features are computed from the original and the transformed versions of RR-interval and HR signals along with a set of direct features that are obtained from ECG signals themselves. In all, 47 features are computed and subsequently they are fed to an ensemble system of bagged decision trees for classifying the ECG recordings into four different classes. The proposed method has been evaluated with 2017 PhysioNet/CinC challenge hidden test dataset (phase II subset) and the final F1 score of 0.81 is obtained.

Bagged Decision Trees Based Scheme of Microgrid Protection Using Windowed Fast Fourier and Wavelet Transforms

Microgrids of varying size and applications are regarded as a key feature of modernizing the power system. The protection of those systems, however, has become a major challenge and a popular research topic for the reason that it involves greater complexity than traditional distribution systems. This paper addresses the issue through a novel approach which utilizes detailed analysis of current and voltage waveforms through windowed fast Fourier and wavelet transforms. The fault detection scheme involves bagged decision trees which use input features extracted from the signal processing stage and selected by correlation analysis. The technique was tested on a microgrid model developed using PSCAD/EMTDS, which is inspired from an operational microgrid in Goldwind Sc. Tech. Co. Ltd, in Beijing, China. The results showed great level of effectiveness to accurately identify faults from other non-fault disturbances, precisely locate the fault and trigger opening of the right circuit breaker/s under different operation modes, fault resistances and other system disturbances.