Training Sample Set Research Articles

A novel sparse polynomial chaos expansion technique with high adaptiveness for surrogate modelling

In this paper, a novel adaptive modelling framework for sparse polynomial chaos expansion is proposed, which can automatically determine adequate truncation degree and training sample set simultaneously. Moreover, the curse of dimensionality issue in polynomial chaos expansion, which generally arises in dealing with high input dimension or large truncation degree, can be alleviated to a large extent. In this framework, a new basis selection strategy, which leverages basis expansion, pruning and refinement, is pursued to adaptively select the polynomial terms of proper degree during the modelling process. Besides, an outstanding sequential sampling strategy is adopted to collect samples of high quality and in relatively small quantity for training polynomial chaos expansion model, and a sparse representation method, Bayesian compressive sensing, is employed for regression calculation. To reconcile the sequential sampling and adaptive basis selection in a consistent framework, a stability evaluation process which works in parallel with the sequential sampling process is performed. The performance of the proposed adaptive modelling framework is evaluated on two benchmark functions and a physical model through comparison with an existing adaptive polynomial chaos expansion modelling technique, a sequential sampling-only approach, and several basis adaptivity strategies. Results demonstrate that the proposed method has high adaptiveness in building surrogate models for various problems. It outperforms the existing adaptive polynomial chaos expansion modelling technique in terms of modelling precision and convergence rate, and it has similar performance with the sequential sampling-only method while lessening the burden in regression calculation and enabling adaptive determination of the truncation degree. Limitations of the proposed method are also summarized after comparing it with various basis adaptivity strategies.

Correction of WindSat sea surface wind speed under rain

To improve the accuracy of sea surface wind speed under rain based on data obtained from the U.S. spaceborne fully polarimetric microwave radiometer (WindSat), a wind speed correction model based on a statistical algorithm is constructed. First, the coupled samples from 2003 to 2014 are considered as the training sample set, and the coupled samples from 2015 to 2020 are used as the test sample set. Then, the WindSat sea surface wind speed under rain correction model is constructed. Results show that the WindSat wind speed data is overestimated severely. The higher the wind speed, the larger the wind speed is overestimated. The experimental results reveal that the WindSat wind speed standard deviation of the test samples is 2.6 m/s. However, the wind speed standard deviation decreases to 2.2 m/s after correction.

Hourly Building Energy Consumption Prediction Using a Training Sample Selection Method Based on Key Feature Search

For the management of building operations, hourly building energy consumption prediction (HBECP) is critical. Many factors, such as energy types, expected day intervals, and acquired feature types, significantly impact HBECP. However, the existing training sample selection methods, especially during transitional seasons, are unable to properly adapt to changes in operational conditions. The key feature search selection (KFSS) approach is proposed in this study. This technique ensures a quick response to changes in the parameters of the predicted day while enhancing the model’s accuracy, stability, and generalization. The best training sample set is found dynamically based on the similarity between the feature on the projected day and the historical data, and feature scenario analysis is used to make the most of the acquired data features. The hourly actual data in two years are applied to a major office building in Zhuhai, China as a case study. The findings reveal that, as compared to the original methods, the KFSS method can track daily load well and considerably enhance prediction accuracy. The suggested training sample selection approach can enhance the accuracy of prediction days by 14.5% in spring and 4.9% in autumn, according to the results. The proposed feature search and feature extraction strategy are valuable for enhancing the robustness of data-driven models for HBECP.

A Combing Data-Driven and Model-Driven Methods for Renewable Energy Acceptance Capacity Assessment in Distribution Networks

To accurately improve the balanced new energy distribution network capacity assessment, this paper proposes a combined data-driven and model-driven renewable energy capacity assessment method for distribution networks. Firstly, the distribution generations (DG) output data are pre-processed to construct a scenario training sample set and extract seven types of DG daily output characteristics indicators. Secondly, a scenario generation model based on a conditional generation adversarial network is proposed to realize the coupling of scenario reduction and scenario generation based on K-means++ through daily output characteristics indicators. Finally, the optimization model based on hybrid positive linear programming is proposed for distributed power supply acceptance capacity evaluation of the distribution network with the objective of maximum DG access capacity and full consideration of voltage deviation, branch current, tidal current return, and other operational constraints. It is validated by IEEE 33-bus simulation. The simulation analysis shows that the proposed method can effectively quantify the DG capacity uncertainty compared with the traditional method, and the proposed method has higher computational accuracy and efficiency than the traditional method.

A classification and identification model of extra virgin olive oil adulterated with other edible oils based on pigment compositions and support vector machine

Adulteration identification of extra virgin olive oil (EVOO) is a vital issue in the olive oil industry. In this study, chromatographic fingerprint data of pigments combined with machine learning methodologies were successfully identified and classified EVOO, refined-pomace olive oil (R-POO), rapeseed oil (RO), soybean oil (SO), peanut oil (PO), sunflower oil (SFO), flaxseed oil (FO), corn oil (CO), extra virgin olive oil adulterated with rapeseed oil (EVOO-RO) and extra virgin olive oil adulterated with corn oil (EVOO-CO). Support vector machine (SVM) classification of EVOO, other edible oils, and EVOO adulteration identification achieved 100% accuracy for the training set sample and 94.44% accuracy for the test set sample. As a result, this SVM model could identify effectively the adulteration EVOO with the limit of 1% RO and 1% CO. Therefore, the excellent classification and predictive power of this model indicated pigments could be used as potential markers for identifying EVOO adulteration.

Near infrared spectroscopy and multivariate statistical analysis as rapid tools for the geographical origin assessment of Italian hazelnuts

The geographical origin assessment of Italian hazelnuts is nowadays a relevant topic, aimed at the protection of provenience certificates. Near Infrared (NIR) spectroscopy could be a functional candidate for preventing and fighting illegal activities related to this matrix. The present study focuses on the exploitability of the NIR technique on the ‘hazelnut chain’ (fresh, roasted and paste), against the false origin declaration frauds, mainly concerning some of the best Italian varieties (‘Nocciola Piemonte’, ‘Tonda Gentile Romana’, ‘Mortarella’). 216 spectra were recorded, for a total of n = 144 for the training set, and n = 72 for the validation set, considering fresh (n = 57), roasted (n = 107), and paste (n = 52) hazelnuts as different matrices. The training set sample selection was made according to a Design of Experiment (DoE), that considered diverse factors, such as harvesting year, storage shelf life, and presence of peel. The validation set was composed of blended samples generated by mixing Italian and non-Italian ones, and real samples bought from local markets. Multivariate Statistical Analysis was employed for data handling and elaboration, both unsupervised and supervised models, Principal Component Analysis, and Partial Least Square-Discriminant Analysis were built to simplify, observe, and classify the samples. A variables selection was performed by filtering the most important ones considering the Variable Importance in Projection (VIP) scores. The predictive ability of the technology was evaluated by applying Classification List and Confusion Matrix approaches to a prediction set, providing a fit of the observations of this set into the selected supervised model. The outcomes highlight valuable discrimination between authentic samples (related to two different harvesting year campaigns) with classification accuracy rates between 89 % and 100 %. Promising results about the application on blended and real samples were also obtained, especially as regards fresh and roasted hazelnuts, which presented classification accuracy rates of 81 % and 91 %. Therefore, this analytical technique could play a strategic role in the geographical origin assessment considering it is a rapid, direct, non-destructive, and cost-effective approach.

Comparative study on landslide susceptibility mapping based on unbalanced sample ratio

The Zigui–Badong section of the Three Gorges Reservoir area is used as the research area in this study to research the impact of unbalanced sample sets on Landslide Susceptibility Mapping (LSM) and determine the sample ratio interval with the best performance for different models. We employ 12 LSM factors, five training sample sets with different sample ratios (1:1, 1:2, 1:4, 1:8, and 1:16), and C5.0, Support Vector Machine (SVM), Logistic Regression (LR), and one-dimensional Convolution Neural Network (CNN) models are used to obtain landslide susceptibility index and landslide susceptibility zoning in the study area, respectively. The prediction performance of the model is evaluated by the receiver operating characteristic curve area under the curve value, five statistical methods, and specific category precision. The results show that the CNN, SVM, and LR models in the sample ratio of 1:2 achieve better performance than on the balanced sample set, which indicates the importance of the unbalanced sample set in training the LSM modeling. The C5.0 model is always in a state of overfitting in this study and needs to be further studied. The conclusions put forward in this study help improve the scientificity and reliability of LSM.

GF-1/6 Satellite Pixel-by-Pixel Quality Tagging Algorithm

The Landsat and Sentinel series satellites contain their own quality tagging data products, marking the source image pixel by pixel with several specific semantic categories. These data products generally contain categories such as cloud, cloud shadow, land, water body, and snow. Due to the lack of mid-wave and thermal infrared bands, the accuracy of traditional cloud detection algorithm is unstable when facing Chinese Gaofen-1/6 (GF-1/6) data. Moreover, it is challenging to distinguish clouds from snow. In order to produce GF-1/6 satellite pixel-by-pixel quality tagging data products, this paper builds a training sample set of more than 100,000 image pairs, primarily using Sentinel-2 satellite data. Then, we adopt the Swin Transformer model with a self-attention mechanism for GF-1/6 satellite image quality tagging. Experiments show that the model’s overall accuracy reaches the level of Fmask v4.6 with more than 10,000 training samples, and the model can distinguish between cloud and snow correctly. Our GF-1/6 quality tagging algorithm can meet the requirements of the “Analysis Ready Data (ARD) Technology Research for Domestic Satellite” project.

A New Three-Way Incremental Naive Bayes Classifier

Aiming at the problems of the dynamic increase in data in real life and that the naive Bayes (NB) classifier only accepts or rejects the sample processing results, resulting in a high error rate when dealing with uncertain data, this paper combines three-way decision and incremental learning, and a new three-way incremental naive Bayes classifier (3WD-INB) is proposed. First, the NB classifier is established, and the distribution fitting is carried out according to the minimum residual sum of squares (RSS) for continuous data, so that 3WD-INB can process both discrete data and continuous data, then carry out an incremental learning operation, select the samples with higher data quality according to the confidence of the samples in the incremental training set for incremental learning, solve the problem of data dynamics and filter the poor samples. Then we construct the 3WD-INB classifier and determine the classification rules of the positive, negative and boundary domains of the 3WD-INB classifier, so that the three-way classification of samples can be realized and better decisions can be made when dealing with uncertain data. Finally, five discrete data and five continuous data are selected for comparative experimental analysis with traditional classification methods. The results show that 3WD-INB has high accuracy and recall rate on different types of datasets, and the classification performance is also relatively stable.

Accurate segmentation of head and neck radiotherapy CT scans with 3D CNNs: consistency is key

Objective. Automatic segmentation of organs-at-risk in radiotherapy planning computed tomography (CT) scans using convolutional neural networks (CNNs) is an active research area. Very large datasets are usually required to train such CNN models. In radiotherapy, large, high-quality datasets are scarce and combining data from several sources can reduce the consistency of training segmentations. It is therefore important to understand the impact of training data quality on the performance of auto-segmentation models for radiotherapy. Approach. In this study, we took an existing 3D CNN architecture for head and neck CT auto-segmentation and compare the performance of models trained with a small, well-curated dataset (n = 34) and then a far larger dataset (n = 185) containing less consistent training segmentations. We performed 5-fold cross-validations in each dataset and tested segmentation performance using the 95th percentile Hausdorff distance and mean distance-to-agreement metrics. Finally, we validated the generalisability of our models with an external cohort of patient data (n = 12) with five expert annotators. Main results. The models trained with a large dataset were greatly outperformed by models (of identical architecture) trained with a smaller, but higher consistency set of training samples. Our models trained with a small dataset produce segmentations of similar accuracy as expert human observers and generalised well to new data, performing within inter-observer variation. Significance. We empirically demonstrate the importance of highly consistent training samples when training a 3D auto-segmentation model for use in radiotherapy. Crucially, it is the consistency of the training segmentations which had a greater impact on model performance rather than the size of the dataset used.

Soil salinity simulation based on electromagnetic induction and deep learning

In pursuit of fine modeling and forecasting of water-salt transport in arid zones, this study uses the rapid electromagnetic induction survey as the main tool to establish an empirical relationship between soil volumetric water content, salinity and electrical conductivity based on on-site measurement data. By using deep learning to set up a surrogate model for electromagnetic induction inverse problem, the vertical distribution of soil salinity during salt leaching at non-growth season is revealed. HYDRUS-1D was used to perform inversions of soil hydraulic parameters to obtain accurate numerical solutions. The results show that the surrogate of electromagnetic induction inversion using a deep convolutional encoding-decoding network (NET) with a learning rate of 0.0006 and 1500 sets of training samples has high accuracy, with an R2 of 0.9930 and RMSE of 0.2585 mg/cm3. The HYDRUS inverse modeling was performed using synthetic ECa data, water content and soil salinity observation data, and the combination of ECa and shallow ground (10 cm deep) water content respectively. It was shown that although the accuracy of the water-salt transport model obtained by using ECa data is slightly lower than that of the inversion model using traditional observation data, it can still accurately simulate the water-salt distribution and has the incomparable advantages of being fast, convenient and low-cost, and the accuracy of the model is improved to some extent by incorporating shallow ground water content data into the inversion based on ECa data. Therefore, the water-salt transport simulation framework incorporating soil apparent electrical conductivity proposed in this study can accurately simulate salt transport and can provide strategies and methods for the simulation and prediction of soil water-salt transport using electromagnetic induction technology.

Training Feedforward Neural Networks Using an Enhanced Marine Predators Algorithm

The input layer, hidden layer, and output layer are three models of the neural processors that make up feedforward neural networks (FNNs). Evolutionary algorithms have been extensively employed in training FNNs, which can correctly actualize any finite training sample set. In this paper, an enhanced marine predators algorithm (MPA) based on the ranking-based mutation operator (EMPA) was presented to train FNNs, and the objective was to attain the minimum classification, prediction, and approximation errors by modifying the connection weight and deviation value. The ranking-based mutation operator not only determines the best search agent and elevates the exploitation ability, but it also delays premature convergence and accelerates the optimization process. The EMPA integrates exploration and exploitation to mitigate search stagnation, and it has sufficient stability and flexibility to acquire the finest solution. To assess the significance and stability of the EMPA, a series of experiments on seventeen distinct datasets from the machine learning repository of the University of California Irvine (UCI) were utilized. The experimental results demonstrated that the EMPA has a quicker convergence speed, greater calculation accuracy, higher classification rate, strong stability and robustness, which is productive and reliable for training FNNs.

A Data Augmented Method for Plant Disease Leaf Image Recognition based on Enhanced GAN Model Network

The identification of plant disease leaves based on deep learning is the key to control the development and spread of plant diseases. In this paper, the existing problems of traditional classification and recognition of plant disease leaves and the limitations of deep learning-based plant disease leaf training are analysed. An enhanced GAN model network based on the Wasserstein GAN loss function has been developed to address the limited training images of plant disease leaves. The self-attention layer is added into the self-encoding structure of the generating network. The effectiveness of data generated by the encoder is increased after the self-attention layer is added after the convolution. Finally, the model's training process is stabilised using the depth gradient punishment method. Three types of corn disease photos and 100 health images from the PlantVillage dataset were used as data sets in the experiment. An AWGAN model was applied to generate around 3000 images. Several data improvement techniques were applied to augment the same datasets. Comparative tests are conducted using AlexNet, VGG16, and ResNet18. The results indicate that the proposed AWGAN model is capable of generating sufficient images of maize leaf diseases with apparent lesions, making it a viable solution for data augmentation of plant disease images. The training model's recognition accuracy is significantly increased. The proposed awGAN-based image identification method for plant leaf disease efficiently resolves the over-fitting problem in the small sample training set. The model recognition accuracy in the ResNet18 network achieves 98.4%.

Improving Ship Detection Based on Decision Tree Classification for High Frequency Surface Wave Radar

The traditional constant false alarm rate (CFAR) method, with fixed parameter settings and single noise background calculation, is unable to intelligently catch the current detection background. To improve the performance of the CFAR method, this paper proposes a target detection method based on decision tree classification (DTC) for high-frequency surface wave radar (HFSWR). Firstly, the training sample set and labels are obtained by means of a ship automatic identification system (AIS). Then, feature vector of range dimension, Doppler dimension and range-Doppler (RD) dimension is extracted by way of cell averaging, ordered statistics, censored mean and trimmed mean. Finally, DTC is used to recognize “true” and “false” targets in feature space. Experimental results show that, under the same number of detection targets, the DTC method is superior to traditional CFAR methods, and the accuracy of target detection can be increased by more than 5%.

A Hybrid Model for Predicting Low Oxygen in the Return Air Corner of Shallow Coal Seams Using Random Forests and Genetic Algorithm

In order to better solve the phenomenon of low oxygen in the corner of return airway caused by abnormal gas emission in goaf during shallow coal seam mining, by analyzing the source and reason of low oxygen phenomenon, a prediction model of oxygen concentration in the corner of return airway based on genetic algorithm (GA) and random forest (RF) technology was proposed. The training sample set was established by using the field data obtained from actual monitoring, including the oxygen concentration in the return airway corner, the periodic pressure step distance of the roof, the surface temperature and atmospheric pressure. GA was used to optimize the parameters in the RF model, including trees and leaves in the forest. The results showed that the model prediction error was minimum when the number of trees was 398 and the number of leaves was 1. In addition, GA was used to optimize the number of hidden neurons and the initial weight threshold of the back-propagation neural network (BPNN). In order to verify the superiority of the model, the GA optimized RF and BPNN model are compared with the conventional RF and BPNN model. Analyze the average absolute percentage error (MAPE), root mean square error (RMSE), and average absolute error (MAE) of the prediction data of each model. The results show that the optimized RF prediction model is better than other models in terms of prediction accuracy.

Identification of Pyroptosis-Relevant Signature in Tumor Immune Microenvironment and Prognosis in Skin Cutaneous Melanoma Using Network Analysis.

Pyroptosis is closely related to the programmed death of cancer cells as well as the tumor immune microenvironment (TIME) via the host-tumor crosstalk. However, the role of pyroptosis-related genes as prognosis and TIME-related biomarkers in skin cutaneous melanoma (SKCM) patients remains unknown. We evaluated the expression profiles, copy number variations, and somatic mutations (CNVs) of 27 genes obtained from MSigDB database regulating pyroptosis among TCGA-SKCM patients. Thereafter, we conducted single-sample gene set enrichment analysis (ssGSEA) for evaluating pyroptosis-associated expression patterns among cases and for exploring the associations with clinicopathological factors and prognostic outcome. In addition, a prognostic pyroptosis-related signature (PPRS) model was constructed by performing Cox regression, weighted gene coexpression network analysis (WGCNA), and least absolute shrinkage and selection operator (LASSO) analysis to score SKCM patients. On the other hand, we plotted the ROC and survival curves for model evaluation and verified the robustness of the model through external test sets (GSE22153, GSE54467, and GSE65904). Meanwhile, we examined the relations of clinical characteristics, oncogene mutations, biological processes (BPs), tumor stemness, immune infiltration degrees, immune checkpoints (ICs), and treatment response with PPRS via multiple methods, including immunophenoscore (IPS) analysis, gene set variation analysis (GSVA), ESTIMATE, and CIBERSORT. Finally, we constructed a nomogram incorporating PPRS and clinical characteristics to improve risk evaluation of SKCM. Many pyroptosis-regulated genes showed abnormal expression within SKCM. TP53, TP63, IL1B, IL18, IRF2, CASP5, CHMP4C, CHMP7, CASP1, and GSDME were detected with somatic mutations, among which, a majority displayed CNVs at high frequencies. Pyroptosis-associated profiles established based on pyroptosis-regulated genes showed markedly negative relation to low stage and superior prognostic outcome. Blue module was found to be highly positively correlated with pyroptosis. Later, this study established PPRS based on the expression of 8 PAGs (namely, GBP2, HPDL, FCGR2A, IFITM1, HAPLN3, CCL8, TRIM34, and GRIPAP1), which was highly associated with OS, oncogene mutations, tumor stemness, immune infiltration degrees, IC levels, treatment responses, and multiple biological processes (including cell cycle and immunoinflammatory response) in training and test set samples. Based on our observations, analyzing modification patterns associated with pyroptosis among diverse cancer samples via PPRS is important, which can provide more insights into TIME infiltration features and facilitate immunotherapeutic development as well as prognosis prediction.

PIGNN-CFD: A physics-informed graph neural network for rapid predicting urban wind field defined on unstructured mesh

Urban wind field plays an important role in quantitative assessment of urban environment. Compared to field measurement and wind tunnel experiment, Computational Fluid Dynamics (CFD) simulation having the advantages of low cost, repeatability and reliable precision is becoming a common scheme to model flow field of one fixed urban scenario, but still faces the problems of time-consuming computation and lack of scalability for practical engineering application. This paper proposes PIGNN-CFD, a novel physics-informed graph neural network for rapid predicting urban wind field based on irregular unstructured mesh data of CFD simulation. Specifically, a CFD model employing the unsteady Reynolds-Averaged Navier-Stokes (RANS) equations with the standard k-ε turbulence model, is constructed and then numerically solved by OpenFOAM to simulate urban wind field defined on unstructured mesh. After being validated by publicly available wind tunnel test data provided by the Architectural Institute of Japan (AIJ), the proposed CFD model is employed to build the training and test sample sets of urban wind fields by simulating the wind blowing through various randomly generated small-scale urban scenes. A novel physics-informed graph neural network, both approximating the training data and automatically satisfying the RANS equations, is designed and trained to perform wind field inference on unstructured mesh graph, and then scaled up to predict wind fields of arbitrary large-scale urban scenes. The predicted wind field results of two urban environments at different scales show that the well-generalized PIGNN-CFD model runs 1–2 orders of magnitude faster than the CFD model on which it is trained, while obtaining the consistent computational accuracy.

Shortening the Standard Testing Time for Residual Biogas Potential (RBP) Tests Using Biogas Yield Models and Substrate Physicochemical Characteristics

The residual biogas potential (RBP) test is a procedure to ensure the anaerobic digestion process performance and digestate stability. Standard protocols for RBP require a significant time for sample preparation, characterisation and testing of the rig setup followed by batch experiments of a minimum of 28 days. To reduce the experimental time to obtain the RBP result, four biogas kinetic models were evaluated for their strength of fit for biogas production data from RBP tests. It was found that the pseudo-parallel first-order model and the first-order autoregressive (AR (1)) model provide a high strength of fit and can predict the RBP result with good accuracy (absolute percentage errors < 10%) using experimental biogas production data of 15 days. Multivariate regression with decision trees (DTs) was adopted in this study to predict model parameters for the AR (1) model from substrate physicochemical parameters. The mean absolute percentage error (MAPE) of the predicted AR (1) model coefficients, the constants and the RBP test results at day 28 across DTs with 20 training set samples are 4.76%, 72.04% and 52.13%, respectively. Using five additional data points to perform the leave-one-out cross-validation method, the MAPEs decreased to 4.31%, 59.29% and 45.62%. This indicates that the prediction accuracy of DTs can be further improved with a larger training dataset. A Gaussian Process Regressor was guided by the DT-predicted AR (1) model to provide probability distribution information for the biogas yield prediction.

Retrieval of Volcanic Ash Cloud Base Height Using Machine Learning Algorithms

There are distinct differences between radiation characteristics of volcanic ash and meteorological clouds, and conventional retrieval methods for cloud base height (CBH) of the latter are difficult to apply to volcanic ash without substantial parameterisation and model correction. Furthermore, existing CBH inversion methods have limitations, including the involvement of many empirical formulae and a dependence on the accuracy of upstream cloud products. A machine learning (ML) method was developed for the retrieval of volcanic ash cloud base height (VBH) to reduce uncertainties in physical CBH retrieval methods. This new methodology takes advantage of polar-orbit active remote-sensing data from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), from vertical profile information and from geostationary passive remote-sensing measurements from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) and the Advanced Geostationary Radiation Imager (AGRI) aboard the Meteosat Second Generation (MSG) and FengYun-4B (FY-4B) satellites, respectively. The methodology involves a statistics-based algorithm with hybrid use of principal component analysis (PCA) and one of four ML algorithms including the k-nearest neighbour (KNN), extreme gradient boosting (XGBoost), random forest (RF), and gradient boosting decision tree (GBDT) methods. Eruptions of the Eyjafjallajökull volcano (Iceland) during April-May 2010, the Puyehue-Cordón Caulle volcanic complex (Chilean Andes) in June 2011, and the Hunga Tonga-Hunga Ha’apai volcano (Tonga) in January 2022 were selected as typical cases for the construction of the training and validation sample sets. We demonstrate that a combination of PCA and GBDT performs more accurately than other combinations, with a mean absolute error (MAE) of 1.152 km, a root mean square error (RMSE) of 1.529 km, and a Pearson’s correlation coefficient (r) of 0.724. Use of PCA as an additional process before training reduces feature relevance between input predictors and improves algorithm accuracy. Although the ML algorithm performs well under relatively simple single-layer volcanic ash cloud conditions, it tends to overestimate VBH in multi-layer conditions, which is an unresolved problem in meteorological CBH retrieval.

Rapid, field-deployable method for determination of seafood decomposition by compact mass spectrometry

Seeking alternatives to sensory which can be performed in non-laboratory field environments has been a recent emphasis in regulatory assessment of seafood decomposition. To this end, an international sampling effort produced samples of 23 different seafood products, which were subjected to controlled decomposition of varying levels on ice. This produced samples of each product for a wide range of decomposition states, and these were scored by a sensory expert on a standard 1-100 scale. Samples were then analyzed by a novel technique involving headspace analysis and compact mass spectrometry. The mass spectrometry data was used to create computer models, guided by the sensory data, to generate a calculated score analogous to a sensory score. Models based on a training set of samples were then used to calculate scores for a test set, and accuracy was assessed by comparing these calculated scores to original sensory data. Calculated scores agreed with sensory findings with respect to overall decomposition state for 96.3% of samples (n=547), with seven false positive (1.3%) and 13 false negative (2.4%) findings. Reproducibility was also assessed via triplicate analysis on separate days for low, middle, and high decomposition states. All sample ranges for these replicates were within 20 points, with 89% (n=66) within 15 points, 73% within 10, and 35% within five. Given the performance of the method, along with the small form factor and manageable operation requirements of the instrument as well as simplicity of sample preparation, this appears to offer a potentially useful technique for field testing of seafood decomposition.