Supervised Classification Models Research Articles

Enhancing Bioactive Compound Classification through the Synergy of Fourier-Transform Infrared Spectroscopy and Advanced Machine Learning Methods.

Bacterial infections and resistance to antibiotic drugs represent the highest challenges to public health. The search for new and promising compounds with anti-bacterial activity is a very urgent matter. To promote the development of platforms enabling the discovery of compounds with anti-bacterial activity, Fourier-Transform Mid-Infrared (FT-MIR) spectroscopy coupled with machine learning algorithms was used to predict the impact of compounds extracted from Cynara cardunculus against Escherichia coli. According to the plant tissues (seeds, dry and fresh leaves, and flowers) and the solvents used (ethanol, methanol, acetone, ethyl acetate, and water), compounds with different compositions concerning the phenol content and antioxidant and antimicrobial activities were obtained. A principal component analysis of the spectra allowed us to discriminate compounds that inhibited E. coli growth according to the conventional assay. The supervised classification models enabled the prediction of the compounds' impact on E. coli growth, showing the following values for accuracy: 94% for partial least squares-discriminant analysis; 89% for support vector machine; 72% for k-nearest neighbors; and 100% for a backpropagation network. According to the results, the integration of FT-MIR spectroscopy with machine learning presents a high potential to promote the discovery of new compounds with antibacterial activity, thereby streamlining the drug exploratory process.

Consistent land use and land cover classification across 20 years of various high-resolution images for detecting soil sealing in Murcia, Spain

Artificial human-induced soil sealing has numerous negative consequences. The extent of impervious surfaces is a key indicator of the location and intensity of human activity; however, it is also proof of damage to the natural environment as a result of the sealing and modification of ecosystems. Remote sensing techniques can help detect and monitor changes in land use and cover over an extended period. However, the limited availability of consistent satellite images with high spatiotemporal resolutions covering several decades poses major challenges for achieving high overall classification accuracy. An accurate methodology for the multitemporal detection of artificial land cover classes was developed and applied to a case study of the metropolitan area of Murcia (Spain) with its challenging landscape conditions due to the frequent presence of bare soil. For this purpose, a variety of high-resolution satellite images from SPOT 5, Rapid Eye, and PlanetScope covering a period of 20 years were used. To improve the automated detection of built-up areas, the reflectance values of the images, normalised difference vegetation index (NDVI) and soil adjusted vegetation index (SAVI), and a building surface digital model were used as inputs for the supervised classification model. We applied a random forest algorithm to non-public, high-resolution images in the Google Earth Engine (GEE) as a processing environment to identify eight target land cover classes. The results show that the proposed methodology leads to a substantial improvement, after including the indices and the digital building model, in the overall accuracy (from 93.16 to 95.97%) and in all classes. This improvement was significant for the artificial classes and was particularly noticeable for the built-up areas (from 91.1 to 95.64%) because their confusion with bare soil was considerably reduced. This work demonstrates the effectiveness of the building-surface digital model as a tool for training the classification model, as it reduces uncertainty in confusion with other spectrally similar classes and its applicability to multisource imagery.

Enhanced Mangrove Index: a spectral index for discrimination understorey, nypa, and mangrove trees

Unsupervised classification using vegetation indices has been extensively employed to map mangrove forests using medium-resolution satellite images. However, its capability is restricted to determining the extent of mangroves only. This study introduces a new spectral index called the Enhanced Mangrove Index (EMI) for accurately mapping different components of mangrove vegetation, including mangrove trees, nypa, and understorey. An immediate effort is required to monitor the invasion of nypa and understorey in the mangrove forest of Segara Anakan Lagoon, located in Central Java, Indonesia. This issue may also be prevalent in other mangrove areas worldwide. The development of EMI involved: 1). the analysis of the reflectance exhibited by different types of mangrove vegetation, and 2). The performance of EMI was evaluated by comparing it with spectral indices such as Automated Mangrove Map and Index (AMMI), as well as supervised classification models like Random Forest (RF). The accuracy assessment indicates that the overall accuracy and Kappa coefficient achieved values of 0.87 and 0.84, respectively, surpassing other spectral indices and supervised classification models. AMMI and RF exhibited high overall accuracy, with values of 0.82 and 0.73, respectively. Additionally, they demonstrated a Kappa coefficient of 0.77 and 0.66, respectively.

Detection of Exploit Websites Using Machine Learning and Data Analysis

Abstract: With the growing integration of internet integration into daily life, the prevalence of security threats such as phishing attacks has become a significant concern. Phishing websites aim to illicitly acquire sensitive user information, posing serious risks to online security. In response, machine learning methods have been investigated for the detection of such fraudulent websites. concentrates on developing a robust phishing website detection system, emphasizing efficiency, accuracy, and costeffectiveness. Four supervised classification models, namely K-Nearest Neighbour, Kernel Support Vector Machine, Decision Tree, and Random Forest Classifier, are employed and compared. Among these models, the Random Forest classifier demonstrates superior performance, achieving an impressive accuracy score of 96.82% on the selected dataset

Organ-delimited gene regulatory networks provide high accuracy in candidate transcription factor selection across diverse processes

Organ-specific gene expression datasets that include hundreds to thousands of experiments allow the reconstruction of organ-level gene regulatory networks (GRNs). However, creating such datasets is greatly hampered by the requirements of extensive and tedious manual curation. Here, we trained a supervised classification model that can accurately classify the organ-of-origin for a plant transcriptome. This K-Nearest Neighbor-based multiclass classifier was used to create organ-specific gene expression datasets for the leaf, root, shoot, flower, and seed in Arabidopsis thaliana. A GRN inference approach was used to determine the: i. influential transcription factors (TFs) in each organ and, ii. most influential TFs for specific biological processes in that organ. These genome-wide, organ-delimited GRNs (OD-GRNs), recalled many known regulators of organ development and processes operating in those organs. Importantly, many previously unknown TF regulators were uncovered as potential regulators of these processes. As a proof-of-concept, we focused on experimentally validating the predicted TF regulators of lipid biosynthesis in seeds, an important food and biofuel trait. Of the top 20 predicted TFs, eight are known regulators of seed oil content, e.g., WRI1, LEC1, FUS3. Importantly, we validated our prediction of MybS2, TGA4, SPL12, AGL18, and DiV2 as regulators of seed lipid biosynthesis. We elucidated the molecular mechanism of MybS2 and show that it induces purple acid phosphatase family genes and lipid synthesis genes to enhance seed lipid content. This general approach has the potential to be extended to any species with sufficiently large gene expression datasets to find unique regulators of any trait-of-interest.

Effects of the Construction of Granadilla Industrial Port in Seagrass and Seaweed Habitats Using Very-High-Resolution Multispectral Satellite Imagery

Seagrass and seaweed meadows hold a very important role in coastal and marine ecosystems. However, anthropogenic impacts pose risks to these delicate habitats. This paper analyses the multitemporal impact of the construction of the largest industrial port in the Canary Islands, near the Special Area of Conservation Natura 2000, on Cymodocea nodosa seagrass meadows (sebadales) of the South of Tenerife, in the locality of Granadilla (Canary Islands, Spain). Very-high-resolution WorldView-2 multispectral satellite data were used for the analysis. Specifically, three images were selected before, during, and after the construction of the port (2011, 2014, and 2022, correspondingly). Initially, advanced pre-processing of the images was performed, and then seabed maps were obtained using the machine learning K-Nearest Neighbors (KNN) supervised classification model, discriminating 12 different bottom types in Case-2 complex waters. The maps achieved high-quality metrics with Precision values of 85%, 81%, and 80%, recall of 76%, 77%, and 77%, and F1 scores of 80%, 79%, and 77% for 2011, 2014, and 2022, respectively. The results mainly show that the construction directly affected the seagrass and seaweed habitats. In particular, the impact of the port on the meadows of Cymodocea nodosa, Caulerpa prolifera, and maërl was assessed. The total maërl population was reduced by 1.9 km2 throughout the study area. However, the Cymodocea nodosa population was maintained at the cost of colonizing maërl areas. Furthermore, the port sedimented a total of 0.98 km2 of seabed, especially Cymodocea nodosa and maërl. In addition, it was observed that Caulerpa prolifera was established as a meadow at the entrance of the port, replacing part of the Cymodocea nodosa and maërl areas. As additional results, bathymetric maps were generated from satellite imagery with the Sigmoid model, and the presence of a submarine outfall was, as well, presented.

The fuzzy support vector data description based on tightness for noisy label detection

Machine learning (ML) is an approach driven by data, and as research in machine learning progresses, the issue of noisy labels has garnered widespread attention. Noisy labels can significantly reduce the accuracy of supervised classification models, making it important to address this problem. Therefore, it is a very meaningful task to detect as many noisy labels as possible from the big data. In this study, a new method is proposed for detecting noisy labels in datasets. This method leverages a deep pre-trained network to extract a feature set from the image data first which can extract more accurate data features. Then, a membership degree based on tightness into the support vector data description (SVDD) model named TF-SVDD is introduced to detect noisy data in the dataset. In order to simulate different types of label noise more accurately, we first assumed that the labels of the datasets used were all correct, and in addition constructed the noise set using two method: the density peak noise set and the random noise set. Experimental results demonstrate that the TF-SVDD can effectively detect noisy label data, surpassing traditional support vector data description algorithms and other methods in terms of outlier detection accuracy, with the average accuracy mostly exceeding 50%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}, and even reaching 80%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}. Furthermore, one novel measure called ‘confidence’ is employed to rectify noisy labels in the data. Following the correction of noisy labels, the accuracy of image classification experiences a significant improvement, with the average promotion ratio mostly exceeding 10%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}, and reaching 30%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}.

Chert sourcing using LIBS: The case of Cova del Parco, Spain

AbstractCherts originating from carbonate lacustrine environments have been widely exploited by Upper Palaeolithic hunter‐gatherers in the south of the Pyrenees. Archaeo‐petrographic sourcing studies have identified different potential sources but were unable to distinguish them. This study conducted geochemical characterisation of geological and archaeological lacustrine chert samples using laser‐induced breakdown spectroscopy (LIBS), and applied the supervised classification model partial least squares–discriminant analysis to LIBS spectra. The model was able to discriminate geological sources and provide predictions for the origin of a selection of chert artefacts recovered at the Middle Magdalenian level from Cova del Parco (Lleida, Spain).

Machine Learning-Guided Discovery of AcrB and MexB Efflux Pump Inhibitors.

Multidrug efflux pump is one of the reasons behind the antimicrobial inactivity related to infection caused by Gram-negative pathogens. The inner membrane resistance-nodulation-cell division transporter proteins, AcrB and MexB, in association with outer membrane proteins, TolC and OprM, are responsible for the extrusion of a broad range of substrates, followed by recognizing them. Although various inhibitors were proposed to stop the efflux activity of the transporter protein, none of them had been approved clinically. Our study aims to identify potent inhibitor-like molecules employing supervised classification models trained upon the molecular descriptors of previously known inhibitors. Based on the intrinsic minimum inhibitory concentration (MIC) values of the reported inhibitors, they were classified into highly potent and less potent categories. A total of 10 different classification models were built using various molecular descriptors; among them, support vector machine, Random Forest, AdaBoost, and LightGBM models appeared to deliver promising results with >80% accuracy. These top four models were implemented on a library of 5043 to obtain 8 hit molecules after the multistep filtering process. To assess their activity toward AcrB and MexB, several molecular dynamics simulations of their ligand-bound structures were performed. We also calculated the binding free-energy values and analyzed other structural properties. Mol.3488 of the unknown molecules showed higher binding affinities for both AcrB and MexB. Also, the presence of "pyridopyrimidone" and "benzothiazole" moieties in the molecules and "V"-shaped orientation of ligands inside the deep binding pocket increase the binding affinity, thereby higher inhibitory properties.

Dense Neural Network for Classification of Seafloor Sediment using Backscatter Mosaic Feature

Water transportation plays a vital role in global economic activities, facilitating more than 85% of international trade and serving as a cost-effective and essential means to fulfill the demand for goods and services. Similarly, the Benoa Port, situated in the southern part of Denpasar City, operates in the same manner. By utilizing Multibeam Echo Sounder (MBES) backscatter data, backscatter mosaics can be generated to identify various seafloor sediment types, which consist of rock fragments, minerals, and organic materials. The characteristics of these sediments, such as grain size, density, composition, and others, can be observed. To improve the classification of sediments, the integration of backscatter data and backscatter features, such as ASM (Angular Second Moment), Energy, Contrast, and Correlation, can be employed. Supervised classification models like Dense Neural Network (DNN) can be utilized to accurately determine the types of seafloor sediments. The application of DNN modeling resulted in a training accuracy rate of 88% and a testing accuracy rate of 100%. The accuracy results delineated six distinct sediment types. Notably, sandy silt exhibited the highest distribution, accounting for 49.30%, whereas soft clayey silt registered the lowest distribution at 0.53%, as determined by their respective spatial prevalence.

Metabolites of the gut microbiota may serve as precise diagnostic markers for sarcopenia in the elderly.

Sarcopenia, a disease recognized by the World Health Organization, has posed a great challenge to the world in the current aging society. The vital role of the gut microbiome through the gut-muscle axis in sarcopenia is increasingly recognized. However, the working mechanisms by which the gut microbiota functions have not been fully explored in the multi-omics field. Here, we designed a cross-sectional study that recruited patients (n = 32) with sarcopenia and healthy old adults (n = 31). Diagnosis of sarcopenia was based on the Asian Working Group for Sarcopenia (AWGS) in 2019 criteria. Muscle mass was represented by appendicular skeletal muscle mass measured by using direct segmental multi-frequency bioelectrical impedance and muscle strength was evaluated using the handgrip strength. The Short Physical Performance Battery, the 5-time Chair Stand Test, and the 4-metre Walk Test were used to assess physical performance. Shotgun metagenomic sequencing was used to profile the gut microbiome in order to identify its construction and function. Metabolome based on untargeted metabolomics was applied to describe the features and structure of fecal metabolites. In clinical indexes including triglycerides and high-density lipoprotein cholesterol, we noted a significant decrease in triglycerides (TG) and a significant increase in high-density lipoprotein cholesterol (HDL-C) in patients with sarcopenia. Appendicular skeletal muscle mass of patients with sarcopenia was lower than the health group. Based on intestinal metagenomic and fecal metabolomic profiles, we found that the gut microbiome and metabolome were disturbed in patients with sarcopenia, with significant decreases in bacteria such as Bifidobacterium longum, Bifidobacterium pseudocatenulatum, and Bifidobacterium adolescentis, as well as metabolites such as shikimic acid. Also, we plotted supervised classification models at the species level of gut bacteria (AUC = 70.83-88.33) and metabolites (AUC = 92.23-98.33) based on machine learning, respectively. Based on the gut-muscle axis network, a potential mechanism is proposed along the gut microbiome - key metabolites - clinical index, that Phascolarctobacterium faecium affects appendicular skeletal muscle mass, calf circumference, handgrip strength, and BMI via Shikimic acid metabolites. This study elucidates the potential mechanisms by which the gut microbiome influences the progress of sarcopenia through metabolites and provides a meaningful theoretical foundation for reference in the diagnosis and treatment of sarcopenia.

Identifying Public Tenders of Interest Using Classification Models: A Comparative Analysis

Public procurement processes related to tenders represent one of the most important financing strategies for companies in diverse sectors, as they present the opportunity to offer services, supplies and product sales to state entities. Given the high volume of public tenders in the Colombian Government SECOP II database, manually identifying tenders of interest can be a cumbersome and time-consuming process. In this work, we propose automating the identification of interesting tenders by training a supervised classification model. We manually label a sample of tenders published in the National Government open data platform, according to whether or not they are of interest to the Minuto de Dios University, and use them for model training. Several models are evaluated in order to select the best model for deployment, taking into account various metrics to determine best performance according to business needs. The best model is selected based on different analyses, comparing the application of data balancing techniques, the performance of the proposed models, and hyper-parameter settings measured against the test data.

A data expansion technique based on training and testing sample to boost the detection of SSVEPs for brain-computer interfaces

Objective. Currently, steady-state visual evoked potentials (SSVEPs)-based brain-computer interfaces (BCIs) have achieved the highest interaction accuracy and speed among all BCI paradigms. However, its decoding efficacy depends deeply on the number of training samples, and the system performance would have a dramatic drop when the training dataset decreased to a small size. To date, no study has been reported to incorporate the unsupervised learning information from testing trails into the construction of supervised classification model, which is a potential way to mitigate the overfitting effect of limited samples. Approach. This study proposed a novel method for SSVEPs detection, i.e. cyclic shift trials (CSTs), which could combine unsupervised learning information from test trials and supervised learning information from train trials. Furthermore, since SSVEPs are time-locked and phase-locked to the onset of specific flashes, CST could also expand training samples on the basis of its regularity and periodicity. In order to verify the effectiveness of CST, we designed an online SSVEP-BCI system, and tested this system combined CST with two common classification algorithms, i.e. extended canonical correlation analysis and ensemble task-related component analysis. Main results. CST could significantly enhance the signal to noise ratios of SSVEPs and improve the performance of systems especially for the condition of few training samples and short stimulus time. The online information transfer rate could reach up to 236.19 bits min−1 using 36 s calibration time of only one training sample for each category. Significance. The proposed CST method can take full advantages of supervised learning information from training samples and unsupervised learning information of testing samples. Furthermore, it is a data expansion technique, which can enhance the SSVEP characteristics and reduce dependence on sample size. Above all, CST is a promising method to improve the performance of SSVEP-based BCI without any additional experimental burden.

Transformer based multiple instance learning for WSI breast cancer classification

The computer-aided diagnosis method based on deep learning provides pathologists with preliminary diagnostic opinions and improves their work efficiency. Inspired by the widespread use of transformers in computer vision, we try to explore their effectiveness and potential in classifying breast cancer tissues in WSIs, and propose a hybrid multiple instance learning method called HTransMIL. Specifically, its first stage is to select informative instances based on hierarchical Swin Transformer, which can capture global and local information of pathological images and is beneficial for obtaining accurate discriminative instances. The second stage aims to strengthen the correlation between selected instances via another transformer encoder consistently and produce powerful bag-level features by aggregating interactived instances for classification. Besides, visualization analysis is utilized to better understand the weakly supervised classification model for WSIs. The extensive evaluation results on a private and two public WSI breast cancer datasets demonstrate the effectiveness and competitiveness of HTransMIL. The code and models are publicly available at https://github.com/Chengyang852/Transformer-for-WSI-classification.

EMG-based hand gesture classifier robust to daily variation: Recursive domain adversarial neural network with data synthesis

ObjectiveIf a supervised classification model is used to predict hand gestures using electromyography (EMG), the EMG signals for training should be labeled every day due to their daily variations. However, annotating these signals every day is time-consuming. MethodsTo address this problem, this study proposes a new framework that updates the EMG classifier in a semi-supervised manner; the classifier was optimized to a target day by using the labeled past EMG signals and the unlabeled signals of the target day. Specifically, the following models were integrated to maximize the classification accuracy: first, a domain-adversarial neural network (DANN) was used to account for the domain shift between the EMG signals of the past and target days. Second, the past dataset was augmented by a data synthesis model incorporated with clustering, random selection, and correlation (CRC). Lastly, a recursive DANN structure was developed to augment the unlabeled EMG signals of the target day. ResultsThe performance of the proposed framework was validated with EMG data of four subjects and five different days. The classification accuracies are 58.22% (without DANN), 66.65% (with DANN), and 73.67% (with recursive DANN and CRC), respectively. ConclusionsThe recursive DANN with CRC enhances the domain adaptation effect, and thus it can address the daily variation problem of EMG. SignificanceThe proposed framework can also be extended to resolve a subject variation problem of EMG.

Stable isotopic fingerprinting of authentic basmati rice from Pakistan

Basmati rice is a popular, high-value aromatic rice which is an easy target for fraudulent activities such as mislabeling and substitution with non-basmati cultivars. Bulk isotope ratio mass spectrometry (IRMS) combined with multivariate analysis was employed to develop a model to identify the geographical origin of rice and authenticate different rice cultivars from Pakistan. ANOVA showed significant statistical differences for δ13C, δ15N, δ2H and δ18O isotopes amongst different basmati and non-basmati rice cultivars. δ2H and δ18O values showed a larger variation between basmati and non-basmati rice cultivars. Multivariate ANOVA showed a significant influence on rice from different regions, cultivars and their stable isotopic values. Regional effects contributed the highest variation for both δ13C and δ15N isotopes (56.6 % and 42.7 %), whereas cultivar type largely contributed to differences in δ2H and δ18O values. Finally, supervised classification models (LDA and PLS-DA) were constructed to assess origin classification ability. The PLS-DA model achieved a higher origin classification accuracy in both training and validation sets (76.7 % and 70.0 %, respectively). In conclusion, isotopic fingerprints along with multivariate statistical analysis showed good potential to characterize rice according to cultivar type and production region. This study provides valuable insights to authenticate Pakistan basmati rice from other non-basmati cultivars, and will be instrumental in developing new tools for industry and regulatory agencies to control fraudulent labeling of basmati rice.

Generating collective counterfactual explanations in score-based classification via mathematical optimization

Due to the increasing use of Machine Learning models in high stakes decision making settings, it has become increasingly important to have tools to understand how models arrive at decisions. Assuming an already trained Supervised Classification model, post-hoc explanations can be obtained via so-called counterfactual analysis: a counterfactual explanation of an instance indicates how this instance should be minimally modified so that the perturbed instance is classified in the desired class by the given Machine Learning classification model.Most of the Counterfactual Analysis literature focuses on the single-instance single-counterfactual setting, in which the analysis is done for one single instance to provide one single counterfactual explanation. Taking a stakeholder’s perspective, in this paper we introduce the so-called collective counterfactual explanations. By means of novel Mathematical Optimization models, we provide a counterfactual explanation for each instance in a group of interest, so that the total cost of the perturbations is minimized under some linking constraints. Making the process of constructing counterfactuals collective instead of individual enables us to detect the features that are critical to the entire dataset to have the individuals classified in the desired class. Our methodology allows for some instances to be treated individually, as in the single-instance single-counterfactual case, performing the collective counterfactual analysis for a fraction of records of the group of interest. This way, outliers are identified and handled appropriately.Under some assumptions on the classifier and the space in which counterfactuals are sought, finding collective counterfactual explanations is reduced to solving a convex quadratic linearly constrained mixed integer optimization problem, which, for datasets of moderate size, can be solved to optimality using existing solvers.The performance of our approach is illustrated on real-world datasets, demonstrating its usefulness.

A high-performance voting-based ensemble model of graph convolutional extreme learning machines for identifying geochemical anomalies related to mineralization

How to effectively identify mineralization anomalies from geochemical survey data is one of the key problems that must be properly solved in geochemical exploration. Although supervised machine learning and deep learning techniques have been adopted to solve this problem, the performance of the established mineralization anomaly identification models is not satisfactory because the mutual relationship between neighborhood data points is not taken into account in the modeling of geochemical exploration data. The graph convolutional extreme learning machines (GCELMs) provide a new technique for modeling the graph data transformed from geochemical exploration data. To improve the performance of mineralization anomaly identification algorithms, the GCELM techniques were adopted to establish the supervised classification models for identifying mineralization anomalies from geochemical exploration data. To improve the robustness of the GCELM models in mineralization anomaly identification, an ensemble classification model called voting-based GCELM (V-GCELM) model was established by voting the outputs of multiple GCELM models. The Baishan area of Jilin Province (China) was taken as the study area for carrying out an experiment of mineralization anomaly identification from geochemical exploration data. The synthetic minority over-sampling technique (SMOTE) was used to alleviate the class-imbalance of the training data. The GCELM model, V-GCELM model and extreme learning machine (ELM) model were established for identifying mineralization anomalies, and their performances were compared in terms of the receiver operating characteristic (ROC) curves, area under the curves (AUCs) and lift indices. The results show that the V-GCELM model has the best performance, and its AUC and lift index are 0.9989 and 28.6, respectively. The GCELM model has the second-best performance, and its AUC and lift index are 0.9977 and 20.1, respectively. The ELM model has the worst but still high performance, its AUC and lift index are 0.9402 and 9.2, respectively. The mineralization anomalies identified by the three models have close spatial correlation with the polymetallic deposits found in the study area, and spatially controlled by the main controlling factors in the study area. Therefore, GCELM and V-GCELM are promising techniques for building the high-performance classification models for identifying mineralization anomalies from geochemical exploration data.

Who's your data? Primary immune deficiency differential diagnosis prediction via machine learning and data mining of the USIDNET registry

PurposeThere are currently more than 480 primary immune deficiency (PID) diseases and about 7000 rare diseases that together afflict around 1 in every 17 humans. Computational aids based on data mining and machine learning might facilitate the diagnostic task by extracting rules from large datasets and making predictions when faced with new problem cases. In a proof-of-concept data mining study, we aimed to predict PID diagnoses using a supervised machine learning algorithm based on classification tree boosting. MethodsThrough a data query at the USIDNET registry we obtained a database of 2396 patients with common diagnoses of PID, including their clinical and laboratory features. We kept 286 features and all 12 diagnoses to include in the model. We used the XGBoost package with parallel tree boosting for the supervised classification model, and SHAP for variable importance interpretation, on Python v3.7. The patient database was split into training and testing subsets, and after boosting through gradient descent, the predictive model provides measures of diagnostic prediction accuracy and individual feature importance. After a baseline performance test, we used the Class Weighting Hyperparameter, or scale_pos_weight to correct for imbalanced classification. ResultsThe twelve PID diagnoses were CVID (1098 patients), DiGeorge syndrome, Chronic granulomatous disease, Congenital agammaglobulinemia, PID not otherwise classified, Specific antibody deficiency, Complement deficiency, Hyper-IgM, Leukocyte adhesion deficiency, ectodermal dysplasia with immune deficiency, Severe combined immune deficiency, and Wiskott-Aldrich syndrome. For CVID, the model found an accuracy on the train sample of 0.80, with an area under the ROC curve (AUC) of 0.80, and a Gini coefficient of 0.60. In the test subset, accuracy was 0.76, AUC 0.75, and Gini 0.51. The positive feature value to predict CVID was highest for upper respiratory infections, asthma, autoimmunity and hypogammaglobulinemia. Features with the highest negative predictive value were high IgE, growth delay, abscess, lymphopenia, and congenital heart disease. For the rest of the diagnoses, accuracy stayed between 0.75 and 0.99, AUC 0.46–0.87, Gini 0.07–0.75, and LogLoss 0.09–8.55. DiscussionClinicians should remember to consider the negative predictive features together with the positives. We are calling this a proof-of-concept study to continue with our explorations. A good performance is encouraging, and feature importance might aid feature selection for future endeavors. In the meantime, we can learn from the rules derived by the model and build a user-friendly decision tree to generate differential diagnoses.

Osteoporosis screening using machine learning and electromagnetic waves

Osteoporosis is a disease characterized by impairment of bone microarchitecture that causes high socioeconomic impacts in the world because of fractures and hospitalizations. Although dual-energy X-ray absorptiometry (DXA) is the gold standard for diagnosing the disease, access to DXA in developing countries is still limited due to its high cost, being present only in specialized hospitals. In this paper, we analyze the performance of Osseus, a low-cost portable device based on electromagnetic waves that measures the attenuation of the signal that crosses the medial phalanx of a patient’s middle finger and was developed for osteoporosis screening. The analysis is carried out by predicting changes in bone mineral density using Osseus measurements and additional common risk factors used as input features to a set of supervised classification models, while the results from DXA are taken as target (real) values during the training of the machine learning algorithms. The dataset consisted of 505 patients who underwent osteoporosis screening with both devices (DXA and Osseus), of whom 21.8% were healthy and 78.2% had low bone mineral density or osteoporosis. A cross-validation with k-fold = 5 was considered in model training, while 20% of the whole dataset was used for testing. The obtained performance of the best model (Random Forest) presented a sensitivity of 0.853, a specificity of 0.879, and an F1 of 0.859. Since the Random Forest (RF) algorithm allows some interpretability of its results (through the impurity check), we were able to identify the most important variables in the classification of osteoporosis. The results showed that the most important variables were age, body mass index, and the signal attenuation provided by Osseus. The RF model, when used together with Osseus measurements, is effective in screening patients and facilitates the early diagnosis of osteoporosis. The main advantages of such early screening are the reduction of costs associated with exams, surgeries, treatments, and hospitalizations, as well as improved quality of life for patients.