Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Feature Extraction”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 181 records · Page 10
All sourcesNASA NTRSDOE OSTI
DOE OSTI2019

Machine Learning Prediction of H Adsorption Energies on Ag Alloys

Adsorption energies on surfaces are excellent descriptors of their chemical properties, including their catalytic performance. This has prompted a significant amount of work towards efficiently predicting adsorption energies without performing expensive quantum chemical calculations, based on semi-empirical or data scientific models. Here, we develop machine learning models to predict H adsorption energies on stepped Ag surface alloys using a database of energies obtained from first-principles electronic structure calculations based on density functional theory. Starting with a simple model and iteratively refining the features extracted from the calculations, we determine the key characteristics of successful models, and reduce the number of features required for accurate prediction of the adsorption energy. We find that nonlinear models are essential for accurate predictions. Moreover, detailed comparison of model predictions and calculations reveal a counterintuitive "selfpassivation" induced by some alloying elements, as well as unexpected discontinuous electronic structure effects. These findings are rationalized using tight-binding model hamiltonians and by analyzing the electronic density of states. Lastly, we show that we can estimate the reliability of a particular prediction, mitigating the effects of errors in the small fraction of cases where these errors are significant.

36 MATERIALS SCIENCE↗
DOE OSTI2015

Morpho-Hawk

Morpho-Hawk applies projectional mathematical morphology in a fundamentally new way. Morpho-Hawk extracts object information from digital images by filtering with simple geometrical figures such as rectangles or simple curves. Two core algorithms are used to accomplish this: 1) Object Detection and Feature Extraction from Images: Using projectional morphology, Morpho-Hawk finds features of interest within an image, such as contours, shapes, colors, infrared spectra, and more. Because Morpho-Hawk identifies features based on shape or form, the method can uniquely handle different image or object conditions and directions. MorphoHawk can evaluate all possible images of the analyzed scene using the same transformations that are applied just to one image of that scene. Hence, recognizing a defect within an image or part of an image provides information to assist in recognizing other objects in the image. In addition, known background image information can be morphologically subtracted out in order to focus on the appearance or disappearance of the features of interest. 2) Object Analysis: Upon detection of an object of interest, Morpho-Hawk can analyze the object based learned information from prior images or end-user defined criteria for certain object of special interest by selecting feature (e.g., color, size, shape, apparent volume) and/or symbol (e.g., letter, number). The novel algorithm can analyze targeted objects, even as conditions of such as illumination, shadows, and spectral regions change strongly in following images as compared to the original image. This enables robust recognition, interpolation and prediction. As the analysis shows the presence of these features in the optical signal, the algorithm can make a determination based on user defined probability. The object information can be stored for later analysis using simple morphological data compression methods and/or retention of the original optical images.

Glazoff, MichaelV.↗
DOE OSTI2016

Interobserver variability in identification of breast tumors in MRI and its implications for prognostic biomarkers and radiogenomics

Purpose: To assess the interobserver variability of readers when outlining breast tumors in MRI, study the reasons behind the variability, and quantify the effect of the variability on algorithmic imaging features extracted from breast MRI. Methods: Four readers annotated breast tumors from the MRI examinations of 50 patients from one institution using a bounding box to indicate a tumor. All of the annotated tumors were biopsy proven cancers. The similarity of bounding boxes was analyzed using Dice coefficients. An automatic tumor segmentation algorithm was used to segment tumors from the readers’ annotations. The segmented tumors were then compared between readers using Dice coefficients as the similarity metric. Cases showing high interobserver variability (average Dice coefficient <0.8) after segmentation were analyzed by a panel of radiologists to identify the reasons causing the low level of agreement. Furthermore, an imaging feature, quantifying tumor and breast tissue enhancement dynamics, was extracted from each segmented tumor for a patient. Pearson’s correlation coefficients were computed between the features for each pair of readers to assess the effect of the annotation on the feature values. Finally, the authors quantified the extent of variation in feature values caused by each of the individual reasons for low agreement. Results: The average agreement between readers in terms of the overlap (Dice coefficient) of the bounding box was 0.60. Automatic segmentation of tumor improved the average Dice coefficient for 92% of the cases to the average value of 0.77. The mean agreement between readers expressed by the correlation coefficient for the imaging feature was 0.96. Conclusions: There is a moderate variability between readers when identifying the rectangular outline of breast tumors on MRI. This variability is alleviated by the automatic segmentation of the tumors. Furthermore, the moderate interobserver variability in terms of the bounding box does not translate into a considerable variability in terms of assessment of enhancement dynamics. The authors propose some additional ways to further reduce the interobserver variability.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2021

Deep Cellular Recurrent Network for Efficient Analysis of Time-Series Data With Spatial Information

Efficient processing of large-scale time series data is an intricate problem in machine learning. Conventional sensor signal processing pipelines with hand engineered feature extraction often involve huge computational cost with high dimensional data. Deep recurrent neural networks have shown promise in automated feature learning for improved time-series processing. However, generic deep recurrent models grow in scale and depth with increased complexity of the data. This is particularly challenging in presence of high dimensional data with temporal and spatial characteristics. Consequently, this work proposes a novel deep cellular recurrent neural network (DCRNN) architecture to efficiently process complex multi-dimensional time series data with spatial information. Here, the cellular recurrent architecture in the proposed model allows for location-aware synchronous processing of time series data from spatially distributed sensor signal sources. Extensive trainable parameter sharing due to cellularity in the proposed architecture ensures efficiency in the use of recurrent processing units with high-dimensional inputs. This study also investigates the versatility of the proposed DCRNN model for classification of multi-class time series data from different application domains. Consequently, the proposed DCRNN architecture is evaluated using two time-series datasets: a multichannel scalp EEG dataset for seizure detection, and a machine fault detection dataset obtained in-house. The results suggest that the proposed architecture achieves state-of-the-art performance while utilizing substantially less trainable parameters when compared to comparable methods in the literature.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

SU-D-202-02: Quantitative Imaging: Correlation Between Image Feature Analysis and the Accuracy of Manually Drawn Contours On PET Images

Purpose: To evaluate the correlation between image features and the accuracy of manually drawn target contours on synthetic PET images Methods: A digital PET phantom was used in combination with Monte Carlo simulation to create a set of 26 simulated PET images featuring a variety of tumor shapes and activity heterogeneity. These tumor volumes were used as a gold standard in comparisons with manual contours delineated by 10 radiation oncologist on the simulated PET images. Metrics used to evaluate segmentation accuracy included the dice coefficient, false positive dice, false negative dice, symmetric mean absolute surface distance, and absolute volumetric difference. Image features extracted from the simulated tumors consisted of volume, shape complexity, mean curvature, and intensity contrast along with five texture features derived from the gray-level neighborhood difference matrices including contrast, coarseness, busyness, strength, and complexity. Correlation between these features and contouring accuracy were examined. Results: Contour accuracy was reasonably well correlated with a variety of image features. Dice coefficient ranged from 0.7 to 0.90 and was correlated closely with contrast (r=0.43, p=0.02) and complexity (r=0.5, p<0.001). False negative dice ranged from 0.10 to 0.50 and was correlated closely with contrast (r=0.68, p<0.001) and complexity (r=0.66, p<0.001). Absolute volumetric difference ranged from 0.0002 to 0.67 and was correlated closely with coarseness (r=0.46, p=0.02) and complexity (r=0.49, p=0.008). Symmetric mean absolute difference ranged from 0.02 to 1 and was correlated closely with mean curvature (r=0.57, p=0.02) and contrast (r=0.6, p=0.001). Conclusion: The long term goal of this study is to assess whether contouring variability can be reduced by providing feedback to the practitioner based on image feature analysis. The results are encouraging and will be used to develop a statistical model which will enable a prediction of contour accuracy based purely on image feature analysis.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2023

Localized keyhole pore prediction during laser powder bed fusion via multimodal process monitoring and X-ray radiography

Systematic fault detection and control during laser powder bed fusion (L-PBF) has been a long-standing objective for system manufacturers and researchers in the additive manufacturing (AM) industry. This manuscript investigates a data fusion approach for detection of keyhole porosity formation during laser irradiation of Ti-6Al-4V substrates by concurrent recording of thermally induced optical emission measured using both off-axis and coaxial photodiode sensors, and acoustic emission. Subsurface defect formation was monitored via high-speed synchrotron X-ray imaging at 20,000 frames per second, enabling temporal registration of keyhole pore formation events to the monitoring signals at a resolution of 50 µs. We developed data fusion machine learning (ML) models for localized prediction of keyhole pore formation at various time scales ranging from 0.5 ms to 2 ms. The signal segments were featurized using two independent approaches: (1) power spectral density (PSD) and (2) highly comparative time series analysis (HCTSA) framework. The extracted features from different sensor modalities were fused together to construct a multimodal feature space and sequential feature selection was used to determine the most informative features for training the ML models. The predictive performance was evaluated for three classifying algorithms: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Gaussian Naive Bayes (GNB). As a result, pore formation events were predicted with up to 0.95 F1-score, 1.0 recall and 0.94 accuracy. The most heavily weighted features indicate that model performance is chiefly governed by the acoustic monitoring signal, with a secondary contribution from the optical emission sensors.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Active Learning Surrogates for Integrating Electron Microscopy and Computational Insights from Simulations in Autonomous Experiments

Artificial Intelligence (AI) combined with simulations and experiments has great potential to accelerate scientific discovery across technology and pharmaceuticals. However, the gap between simulations and experiments is challenging due to disparities in time and scale, making it difficult to estimate properties like energy and electronic states from experiments, and to provide feedback based on theoretical insights.Our research addresses the challenge by developing unique deep kernel based surrogate models that learns from microscopic images, mapping structural features to energy differences from defect formation. We start with full-training using simulated images to determine optimal settings, establishing a baseline for active learning. Using these settings from the baseline, active learning is trained, and predicts structures along simulation trajectories based on uncertainty and energetic stability, thus reducing data requirements, simulation time and computational costs. The results demonstrate that the model achieves a low average error margin of approximately 0.03 meV, indicating good performance. To enhance feature extraction and reconstruction capabilities, we developed an autoencoder-decoder as additional surrogate to create latent space to capture essential features, enabling precise comparisons between simulations and experiments. The results from this model achieved a reconstruction loss of around 0.2 and accurately reconstructed molecular structures.Overall, this work advances the steering of experiments through computational simulations by employing a surrogate models that actively predicts the trajectories of structural evolution, achieving time-to-solution comparable to experimental measurements.

Saranathan, Gayathri [Hewlett-Packard]↗
DOE OSTI2019

LANSCE Diagnostic Robot Localization

Los Alamos Neutron Science Center (LANSCE) operates a linear particle accelerator (LINAC) that is used for a number of scientific research projects. Due to the presence of harmful radiation, researchers are not allowed in the beam tunnel during operation. Since the beam tunnel is inaccessible, an autonomous mobile robot is to be developed and deployed in the tunnel, monitoring the accelerator during operation. The robot will present real time data for operators and scientists from sensors such as a video camera, thermal camera, microphones, and muon detectors; allowing for beam diagnostics previously unavailable. Localization is a fundamental step in autonomous navigation of mobile robots, answering the “where am I” question. For this project, an Extended Kalman Filter (EKF) algorithm is implemented as a positional estimator. The EKF relies on an odometric motion model to predict the robot’s position and LIDAR measurement data to update the robot’s position. The measurement model is based on v features extracted from the LIDAR point cluster using the Split-and-Merge algorithm, and the nearest neighbor algorithm associates the features to an a priori feature map of the beam tunnel. The predicted and measured positions are then joined using a weighted value by means of the Kalman gain. The effectiveness of this localization technique is demonstrated using the LabVIEW robotics simulator.

42 ENGINEERING↗
DOE OSTI2021

A meta-learning based distribution system load forecasting model selection framework

This paper presents a meta-learning based, automatic distribution system load forecasting model selection framework. Furthermore, the framework includes the following processes: feature extraction, candidate model preparation and labeling, offline training, and online model recommendation. Using load forecasting needs and data characteristics as input features, multiple metalearners are used to rank the candidate load forecast models based on their forecasting accuracy. Then, a scoring-voting mechanism is proposed to weights recommendations from each meta-leaner and make the final recommendations. Heterogeneous load forecasting tasks with different temporal and technical requirements at different load aggregation levels are set up to train, validate, and test the performance of the proposed framework. Simulation results demonstrate that the performance of the meta-learning based approach is satisfactory in both seen and unseen forecasting tasks.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2022

SNAD transient miner: Finding missed transient events in ZTF DR4 using k-D trees

Here we report the automatic detection of 11 transients (7 possible supernovae and 4 active galactic nuclei candidates) within the Zwicky Transient Facility fourth data release (ZTF DR4), all of them observed in 2018 and absent from public catalogs. Among these, three were not part of the ZTF alert stream. Our transient mining strategy employs 41 physically motivated features extracted from both real light curves and four simulated light curve models (SN Ia, SN II, TDE, SLSN-I). These features are input to a k-D tree algorithm, from which we calculate the 15 nearest neighbors. After pre-processing and selection cuts, our dataset contained approximately a million objects among which we visually inspected the 105 closest neighbors from seven of our brightest, most well-sampled simulations, comprising 89 unique ZTF DR4 sources. Our result illustrates the potential of coherently incorporating domain knowledge and automatic learning algorithms, which is one of the guiding principles directing the SNAD team. It also demonstrates that the ZTF DR is a suitable testing ground for data mining algorithms aiming to prepare for the next generation of astronomical data.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Exploiting Multi-Domain Features for Detection of Unclassified Electromagnetic Signals

Deep Learning based classification techniques have shown excellent performance in static environments, where the training and testing samples are drawn from the same distribution. However, real world scenarios often present samples that do not belong to the known set of classes chosen during training. This is quite common for electromagnetic signals, where it is impractical to assume that all possible waveforms are known a-priori, specially in scenarios like warfare. To address this problem, we propose a deep learning based adversarial model where the generator learns to generate waveform features that can deceive the discriminator model as true samples. We introduce domain knowledge of wireless signals by decomposing the signal into a lower dimensional unique feature set, which is used for classifying known versus unknown signals. We further introduce multiple domain representations of the signal to extract features and combine them together to accurately classify new waveforms as an unknown class. Our results show that combined features from multiple domains outperform any single domain representation, especially at low SNR regimes with fewer number of samples to classify.

99 - GENERAL AND MISCELLANEOUS↗
DOE OSTI2024

Exploiting Multi-Domain Features for Detection of Unclassified Electromagnetic Signals (Presentation)

Deep Learning based classification techniques have shown excellent performance in static environments, where the training and testing samples are drawn from the same distribution. However, real world scenarios often present samples that do not belong to the known set of classes chosen during training. This is quite common for electromagnetic signals, where it is impractical to assume that all possible waveforms are known a-priori, specially in scenarios like warfare. To address this problem, we propose a deep learning based adversarial model where the generator learns to generate waveform features that can deceive the discriminator model as true samples. We introduce domain knowledge of wireless signals by decomposing the signal into a lower dimensional unique feature set, which is used for classifying known versus unknown signals. We further introduce multiple domain representations of the signal to extract features and combine them together to accurately classify new waveforms as an unknown class. Our results show that combined features from multiple domains outperform any single domain representation, especially at low SNR regimes with fewer number of samples to classify.

99 - GENERAL AND MISCELLANEOUS↗
DOE OSTI2022

Improving Single-Stage Object Detectors for Nighttime Pedestrian Detection

We report Improving the reliability of nighttime pedestrian detection is a crucial challenge towards the design of robust autonomous systems. Not surprisingly, most pedestrian fatalities occur in low-illumination settings, thus emphasizing the need for new algorithmic advances. This work presents a novel pedestrian detection approach that makes a number of crucial modifications to the state-of-the-art YOLOV5-PANet architecture, in order to improve the reliability of features extracted from nighttime images. More specifically, the proposed architecture systematically incorporates powerful shuffle attention mechanisms and a transformer module to improve the feature learning pipeline. Instead of advocating the use of other sensing modalities that are better suited for nighttime detection, our approach relies only on conventional RGB cameras and is hence broadly applicable. Our empirical studies with nighttime pedestrian detection benchmarks show that with only minimal increase in model complexity, our approach provides significant improvements in detection efficacy over existing solutions. Finally, we explore the impact of post-hoc network pruning on the speed-accuracy trade-off of our approach and demonstrate that it is well suited for reduced memory/compute requirements.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Feature Selection Techniques for a Machine Learning Model to Detect Autonomic Dysreflexia

Feature selection plays a crucial role in the development of machine learning algorithms. Understanding the impact of the features on a model, and their physiological relevance can improve the performance. This is particularly helpful in the healthcare domain wherein disease states need to be identified with relatively small quantities of data. Autonomic Dysreflexia (AD) is one such example, wherein mismanagement of this neurological condition could lead to severe consequences for individuals with spinal cord injuries. We explore different methods of feature selection needed to improve the performance of a machine learning model in the detection of the onset of AD. We present different techniques used as well as the ideal metrics using a dataset of thirty-six features extracted from electrocardiograms, skin nerve activity, blood pressure and temperature. The best performing algorithm was a 5-layer neural network with five relevant features, which resulted in 93.4% accuracy in the detection of AD. The techniques in this paper can be applied to a myriad of healthcare datasets allowing forays into deeper exploration and improved machine learning model development. Through critical feature selection, it is possible to design better machine learning algorithms for detection of niche disease states using smaller datasets.

electrocardiography↗
DOE OSTI2015

Framework for 3D histologic reconstruction and fusion with in vivo MRI: Preliminary results of characterizing pulmonary inflammation in a mouse model

Purpose: Pulmonary inflammation is associated with a variety of diseases. Assessing pulmonary inflammation on in vivo imaging may facilitate the early detection and treatment of lung diseases. Although routinely used in thoracic imaging, computed tomography has thus far not been compellingly shown to characterize inflammation in vivo. Alternatively, magnetic resonance imaging (MRI) is a nonionizing radiation technique to better visualize and characterize pulmonary tissue. Prior to routine adoption of MRI for early characterization of inflammation in humans, a rigorous and quantitative characterization of the utility of MRI to identify inflammation is required. Such characterization may be achieved by considering ex vivo histology as the ground truth, since it enables the definitive spatial assessment of inflammation. In this study, the authors introduce a novel framework to integrate 2D histology, ex vivo and in vivo imaging to enable the mapping of the extent of disease from ex vivo histology onto in vivo imaging, with the goal of facilitating computerized feature analysis and interrogation of disease appearance on in vivo imaging. The authors’ framework was evaluated in a preclinical preliminary study aimed to identify computer extracted features on in vivo MRI associated with chronic pulmonary inflammation. Methods: The authors’ image analytics framework first involves reconstructing the histologic volume in 3D from individual histology slices. Second, the authors map the disease ground truth onto in vivo MRI via coregistration with 3D histology using the ex vivo lung MRI as a conduit. Finally, computerized feature analysis of the disease extent is performed to identify candidate in vivo imaging signatures of disease presence and extent. Results: The authors evaluated the framework by assessing the quality of the 3D histology reconstruction and the histology—MRI fusion, in the context of an initial use case involving characterization of chronic inflammation in a mouse model. The authors’ evaluation considered three mice, two with an inflammation phenotype and one control. The authors’ iterative 3D histology reconstruction yielded a 70.1% ± 2.7% overlap with the ex vivo MRI volume. Across a total of 17 anatomic landmarks manually delineated at the division of airways, the target registration error between the ex vivo MRI and 3D histology reconstruction was 0.85 ± 0.44 mm, suggesting that a good alignment of the ex vivo 3D histology and ex vivo MRI had been achieved. The 3D histology-in vivo MRI coregistered volumes resulted in an overlap of 73.7% ± 0.9%. Preliminary computerized feature analysis was performed on an additional four control mice, for a total of seven mice considered in this study. Gabor texture filters appeared to best capture differences between the inflamed and noninflamed regions on MRI. Conclusions: The authors’ 3D histology reconstruction and multimodal registration framework were successfully employed to reconstruct the histology volume of the lung and fuse it with in vivo MRI to create a ground truth map for inflammation on in vivo MRI. The analytic platform presented here lays the framework for a rigorous validation of the identified imaging features for chronic lung inflammation on MRI in a large prospective cohort.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2017

Feature Selection and Inferential Procedures for Video Data [Slides]

With the rise of electronic and high-dimensional data, new and innovative feature detection and statistical methods are required to perform accurate and meaningful statistical analysis of these datasets that provide unique statistical challenges. In the area of feature detection, much of the recent feature detection research in the computer vision community has focused on deep learning methods, which require large amounts of labeled training data. However, in many application areas, training data is very limited and often difficult to obtain. We develop methods for fast, unsupervised, precise feature detection for video data based on optical flows, edge detection, and clustering methods. We also use pretrained neural networks and interpretable linear models to extract features using very limited training data. In the area of statistics, while high-dimensional data analysis has been a main focus of recent statistical methodological research, much focus has been on populations of high-dimensional vectors, rather than populations of high-dimensional tensors, which are three-dimensional arrays that can be used to model dependent images, such as images taken of the same person or ripped video frames. Our feature detection method is a non-model-based method that fusses information from dense optical flow, raw image pixels, and frame differences to generate detections. Our hypothesis testing methods are based on the assumption that dependent images are concatenated into a tensor that follows a tensor normal distribution, and from this assumption, we derive likelihood-ratio, score, and regression-based tests for one- and multiple-sample testing problems. Our methods will be illustrated on simulated and real datasets. We conclude this report with comments on the relationship between feature detection and hypothesis testing methods.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Machine learning-based multidomain processing for texture-based image segmentation and analysis

Atomic and molecular resolved atomic force microscopy (AFM) images offer unique insights into materials' properties such as local ordering, molecular orientation, and topological defects, which can be used to pinpoint physical and chemical interactions occurring at the surface. Utilizing machine learning for extracting underlying physical parameters increases the throughput of AFM data processing and eliminates inconsistencies intrinsic to manual image analysis, thus enabling the creation of reliable frameworks for qualitative and quantitative evaluation of experimental data. Here, we present a robust and scalable approach to the segmentation of AFM images based on flexible pre-selected classification criteria. The usage of supervised learning and feature extraction allows us to retain the consideration of specific problem-dependent features (such as types of periodical structures observed in the images and the associated numerical parameters: spacing, orientation, etc.). We highlight the applicability of this approach for the segmentation of molecular resolved AFM images based on the crystal orientation of the observed domains, automated selection of boundaries, and collection of relevant statistics. Finally, we outline a general strategy for machine learning-enabled analysis of nanoscale systems exhibiting periodic order that could be applied to any analytical imaging technique

36 MATERIALS SCIENCE↗
DOE OSTI2022

Unstructured clinical notes within the 24 hours since admission predict short, mid & long-term mortality in adult ICU patients

Mortality prediction for intensive care unit (ICU) patients is crucial for improving outcomes and efficient utilization of resources. Accessibility of electronic health records (EHR) has enabled data-driven predictive modeling using machine learning. However, very few studies rely solely on unstructured clinical notes from the EHR for mortality prediction. In this work, we propose a framework to predict short, mid, and long-term mortality in adult ICU patients using unstructured clinical notes from the MIMIC III database, natural language processing (NLP), and machine learning (ML) models. Depending on the statistical description of the patients’ length of stay, we define the short-term as 48-hour and 4-day period, the mid-term as 7-day and 10-day period, and the long-term as 15-day and 30-day period after admission. We found that by only using clinical notes within the 24 hours of admission, our framework can achieve a high area under the receiver operating characteristics (AU-ROC) score for short, mid and long-term mortality prediction tasks. The test AU-ROC scores are 0.87, 0.83, 0.83, 0.82, 0.82, and 0.82 for 48-hour, 4-day, 7-day, 10-day, 15-day, and 30-day period mortality prediction, respectively. We also provide a comparative study among three types of feature extraction techniques from NLP: frequency-based technique, fixed embedding-based technique, and dynamic embedding-based technique. Lastly, we provide an interpretation of the NLP-based predictive models using feature-importance scores.

60 APPLIED LIFE SCIENCES↗