Search

Objective: This study aimed to use the radiomics signatures of a machine learning-based tool to evaluate the prognosis of patients with coronavirus disease 2019 (COVID-19) infection., Methods: The clinical and imaging data of 64 patients with confirmed diagnoses of COVID-19 were retrospectively selected and divided into a stable group and a progressive group according to the data obtained from the ongoing treatment process. Imaging features from whole-lung images from baseline computed tomography (CT) scans were extracted and dimensionality reduction was performed. Support vector machines were used to construct radiomics signatures and to compare differences between the 2 groups. We also compared the differences of signature scores in the clinical, laboratory, and CT image feature subgroups and finally analyzed the correlation between the radiomics features of the constructed signature and the other features including clinical, laboratory, and CT imaging features., Results: The signature has a good classification effect for the stable group and the progressive group, with area under curve, sensitivity, and specificity of 0.833, 80.95%, and 74.42%, respectively. Signature score differences in laboratory and CT imaging features between subgroups were not statistically significant (P>0.05); cough was negatively correlated with GLCM Entropy_angle 90_offset4 (r=-0.578), but was positively correlated with ShortRunEmphhasis_AllDirect_offset4_SD (r=0.454); C-reactive protein was positively correlated with Cluster Prominence_ AllDirect_offset 4_ SD (r=0.47)., Conclusion: The radiomics signature of the whole lung based on machine learning may reveal the changes of lung microstructure in the early stage and help to indicate the progression of the disease., Competing Interests: The authors declare no conflicts of interest., (Copyright © 2020 The Author(s). Published by Wolters Kluwer Health, Inc.)

Accurate estimation of photometric redshifts (photo- z) is crucial in studies of both galaxy evolution and cosmology using current and future large sky surveys. In this study, we employ Random Forest (RF), a machine learning algorithm, to estimate photo- z and investigate the systematic uncertainties affecting the results. Using galaxy flux and colour as input features, we construct a mapping between input features and redshift by using a training set of simulated data, generated from the Hubble Space Telescope Advanced Camera for Surveys (HST-ACS) and COSMOS catalogue, with the expected instrumental effects of the planned China Space Station Telescope (CSST). To improve the accuracy and confidence of predictions, we incorporate inverse variance weighting and perturb the catalogue using input feature errors. Our results show that weighted RF can achieve a photo- z accuracy of |$\rm \sigma _{NMAD}=0.025$| and an outlier fraction of |$\rm \eta =2.045\ \hbox{per cent}$|⁠ , significantly better than the values of |$\rm \sigma _{NMAD}=0.043$| and |$\rm \eta =6.45\ \hbox{per cent}$| obtained by the widely used Easy and Accurate Zphot from Yale (EAZY) software, which uses template-fitting method. Furthermore, we have calculated the importance of each input feature for different redshift ranges and found that the most important input features reflect the approximate position of the break features in galaxy spectra, demonstrating the algorithm's ability to extract physical information from data. Additionally, we have established confidence indices and error bars for each prediction value based on the shape of the redshift probability distribution function, suggesting that screening sources with high confidence can further reduce the outlier fraction. [ABSTRACT FROM AUTHOR]

The question of radio dichotomy in active galactic nuclei (AGNs) is still in debate, even though it has been discussed for more than 40 years. In order to solve this old riddle, we collect a sample of AGNs with optical B band and radio 6 cm wavelength data to analyze the radio loudness log  R. Our results indicate a separation of log  R = 〈1.37 ± 0.02〉 between radio-loud (RL) AGNs and radio-quiet (RQ) AGNs, suggesting the existence of an RL/RQ dichotomy. For the first time, we suggest combining radio luminosity and radio loudness as a double criterion to divide AGNs into RLs and RQs to avoid misclassification problems that may happen in the single-criterion scenario; we propose the double-criterion dividing line log  L 6 cm = −2.7log  R + 44.3 by using a machine-learning method. In addition, the key point of the RL/RQ dichotomy is the origin of radio emission for the two classes, and we suggest the radio emission from RLs and RQs share the same origin, e.g. jets and mini-jets (aborted-jet or outflow), through a correlation study between radio 6 cm luminosity and optical B -band luminosity. [ABSTRACT FROM AUTHOR]

• Machine learning models were proven to be incapable of unexpected situation prediction. • Support vector regression and Kalman filter were combined for train running time prediction. • The Kalman filter substantially improved the performance of the machine learning models. • The proposed method conforms to the timeliness and high accuracy requirements of real-time prediction. This study aims to propose a hybrid model that comprises support vector regression (SVR) and a Kalman filter (KF) to improve the train running time prediction accuracy of machine learning models during railway disruptions. The SVR was trained using offline data, whereas the KF updated the SVR prediction using real-time information. Thus, the hybrid model mitigates the time-consuming online training of machine learning models and their inability to reflect real-time information when using offline training. To obtain a high-performance prediction model, four key SVR parameters were first optimized based on cross-validation. Then, SVR predictions were evaluated using the mean absolute and percentage errors of the test datasets by considering the trains that suffered disruptions. The results from this evaluation show that the SVR notably outperformed other benchmark models but was unable to provide satisfactory predictions under unexpected situations. Next, we applied the KF to update the SVR prediction using real-time information and conducted model performance evaluation of the predictions based on the hybrid model. The corresponding results show that the KF significantly improved the SVR prediction accuracy under unexpected disruption situations. Furthermore, using offline training, along with the KF instead of online training, substantially reduced the computational time. [ABSTRACT FROM AUTHOR]

• An integrated framework is proposed to predict lane-changing risk. • Driving intentions are recognized using LSTM neural network. • LGBM algorithm achieves a higher lane-changing risk prediction accuracy. • Feature importance analysis is conducted using LGBM classifier. Proactive lane-changing (LC) risk prediction can assist driver's LC decision-making to ensure driving safety. However, most previous studies on LC risk prediction did not consider the driver's intention recognition, which made it difficult to guarantee the timeliness and practicability of LC risk prediction. Moreover, the difference in driving risks and its influencing factors between LC to left lane (LCL) and LC to right lane (LCR) have rarely been investigated. To bridge the above research gaps, this study proposes a proactive LC risk prediction framework which integrates the LC intention recognition module and LC risk prediction module. The Long Short-term Memory (LSTM) neural network with time-series input was employed to recognize the driver's LC intention. The Light Gradient Boosting Machine (LGBM) algorithm was then applied to predict the LC risk. Feature importance analysis was lastly conducted to obtain the key features that affect the LC risk. The highD trajectory dataset was used for framework validation. Results show that the recognition accuracy of the driver's LCL, LCR and lane-keeping (LK) intentions based on the proposed LSTM model are 97%, 96% and 97%, respectively. Meanwhile, the LGBM algorithm outperforms other machine learning algorithms in LC risk prediction. The results from feature importance analysis show that the interaction characteristics of the LC vehicle and its preceding vehicle in the current lane have the greatest impact on the LC risk. The proposed framework could potentially be implemented in advanced driver-assistance system (ADAS) or autonomous driving system for improved driving safety. [ABSTRACT FROM AUTHOR]

• Traffic flow states are classified with smartphone sensor data. • The statistical features of vehicle accelerations of only one car are used. • Acceleration and angular acceleration help improve the classification accuracy. • Optimization of the hyper-parameters further reduces the classification error. This study proposes a framework to classify traffic flow states. The framework is capable of processing massive, high-density, and noise-contaminated data sets generated from smartphone sensors. The statistical features of vehicle acceleration, angular acceleration, and GPS speed data, recorded by smartphone software, are analyzed, and then used as input for traffic flow state classification. Data collected by a five-day experiment is used to train and test the proposed model. A total of 747,856 sets of data are generated and used for both traffic flow state classification and sensitivity analysis of input variables. After applying various algorithms to the proposed framework, the study found that acceleration and angular acceleration data can increase the accuracy of traffic flow classification significantly. When the hyper-parameters of the Deep Belief Network model are optimized by the Differential Evolution Grey Wolf Optimizer algorithm, the classification accuracy is further improved. The results have demonstrated the effectiveness of using smartphone sensor data to estimate the traffic flow states and shown that our proposed model outperforms some traditional machine learning methods in traffic flow state classification accuracy. [ABSTRACT FROM AUTHOR]