Highest Cross-validated Accuracy Research Articles

Machine learning that predicts well may not learn the correct physical descriptions of glassy systems

The complexity of glasses makes it challenging to explain their dynamics. Machine learning (ML) has emerged as a promising pathway for understanding glassy dynamics by linking their structural features to rearrangement dynamics. Support vector machine (SVM) was one of the first methods used to detect such correlations. Specifically, a certain output of SVMs trained to predict dynamics from structure, the distance from the separating hyperplane, was interpreted as being linearly related to the activation energy for the rearrangement. By numerical analysis of toy models, we explore under which conditions it is possible to infer the energy barrier to rearrangements from the distance to the separating hyperplane. We observe that such successful inference is possible only under very restricted conditions. Typical tests, such as the apparent Arrhenius dependence of the probability of rearrangement on the inferred energy and the temperature, or high cross-validation accuracy do not guarantee success. Since even in such relatively simple toy models, prediction success of ML models does not necessarily translate into success of learning the underlying physics, we suggest that more careful investigations are needed when such claims are made. For this, we propose practical approaches for measuring the quality of the energy inference and for modifying the inferred model to improve the inference, which should be usable in the context of realistic datasets. Published by the American Physical Society 2024

Machine Learning-Driven Stroke Prediction Using Independent Dataset

The incidence of stroke cases has witnessed a rapid global rise, affecting not only the elderly but also individuals across all age groups. Accurate prediction of stroke occurrence demands the utilization of extensive data pre-processing techniques. Moreover, the automation of early stroke forecasting is crucial to prevent its onset at the initial stage. In this study, stroke prediction models are evaluated to estimate the likelihood of stroke based on various symptoms such as age, gender, pre-existing medical conditions, and social variables. The machine learning techniques employed include Linear Support Vector Classifier, Extreme Gradient Boosting Classifier, Multilayer Perceptron, Adaptive Boosting Classifier, Bootstrap Aggregating Classifier, and Light Gradient-Boosting Machine. The purpose of this paper is to optimize the hyperparameters of machine learning approaches in developing stroke prediction models. The goal was achieved through a comprehensive comparison of three different sampling techniques for handling imbalanced datasets and evaluating their performance by using various metrics. The most effective model is identified, which is the Adaptive Boosting Classifier utilizing the Tomek Links, with a cross-dataset accuracy of 99% which demonstrated a reliable performance and generalization as evidenced by high cross-validation scores and accuracy on an independent dataset. The next stage of this endeavor entails looking into multiple ways to forecast the development of new dangerous diseases such as breast cancer and skin disorders. In the long run, the aim of subsequent work is to build a powerful toolset that is obtainable to all medical practitioners, allowing for the pre-emptive diagnosis of all potentially hazardous illnesses.

Advancing passive BCIs: a feasibility study of two temporal derivative features and effect size-based feature selection in continuous online EEG-based machine error detection.

The emerging integration of Brain-Computer Interfaces (BCIs) in human-robot collaboration holds promise for dynamic adaptive interaction. The use of electroencephalogram (EEG)-measured error-related potentials (ErrPs) for online error detection in assistive devices offers a practical method for improving the reliability of such devices. However, continuous online error detection faces challenges such as developing efficient and lightweight classification techniques for quick predictions, reducing false alarms from artifacts, and dealing with the non-stationarity of EEG signals. Further research is essential to address the complexities of continuous classification in online sessions. With this study, we demonstrated a comprehensive approach for continuous online EEG-based machine error detection, which emerged as the winner of a competition at the 32nd International Joint Conference on Artificial Intelligence. The competition consisted of two stages: an offline stage for model development using pre-recorded, labeled EEG data, and an online stage 3 months after the offline stage, where these models were tested live on continuously streamed EEG data to detect errors in orthosis movements in real time. Our approach incorporates two temporal-derivative features with an effect size-based feature selection technique for model training, together with a lightweight noise filtering method for online sessions without recalibration of the model. The model trained in the offline stage not only resulted in a high average cross-validation accuracy of 89.9% across all participants, but also demonstrated remarkable performance during the online session 3 months after the initial data collection without further calibration, maintaining a low overall false alarm rate of 1.7% and swift response capabilities. Our research makes two significant contributions to the field. Firstly, it demonstrates the feasibility of integrating two temporal derivative features with an effect size-based feature selection strategy, particularly in online EEG-based BCIs. Secondly, our work introduces an innovative approach designed for continuous online error prediction, which includes a straightforward noise rejection technique to reduce false alarms. This study serves as a feasibility investigation into a methodology for seamless error detection that promises to transform practical applications in the domain of neuroadaptive technology and human-robot interaction.

A deep learning approach to classifying flow-induced vibration response regimes of an elliptical cylinder

This study proposes a new approach that leverages deep learning to the study of flow-induced vibration (FIV), specifically to automate flow regime classification and to visualize the transitions between these regimes. Using previously obtained data on the amplitude response of an elastically mounted cylinder as a function of reduced velocity for a range of structural damping ratios, the time trace of the body displacement and fluid driving forces are first converted into a frequency-time representation using continuous wavelet transforms before being input to several pre-trained convolutional neural networks for feature extraction. When utilizing the outputs of each convolutional neural network for regime classification, we found that almost all the machine learning approaches had high cross-validation accuracy that was statistically insignificant from each other. The five best-performing classifiers were then used as an ensemble method, yielding a weighted accuracy of 99.1% on the test data. The FIV response regimes were further investigated by projecting the outputs of the pre-trained convolutional neural networks onto the first three modes identified with principal component analysis (PCA). The PCA plots indicated that, among all the models considered, Xception showed superior capability in delineating distinct locations for different FIV response regimes, based on the lock-in frequency and the presence of harmonics in the driving fluid forces. Moreover, the PCA plots also showed that increasing the structural damping ratio resulted in a diminished disparity in the dynamics of the identified FIV response regimes, leading to a less discernible separation between the regimes in the plots.

Deep learning models reveal replicable, generalizable, and behaviorally relevant sex differences in human functional brain organization

Sex plays a crucial role in human brain development, aging, and the manifestation of psychiatric and neurological disorders. However, our understanding of sex differences in human functional brain organization and their behavioral consequences has been hindered by inconsistent findings and a lack of replication. Here, we address these challenges using a spatiotemporal deep neural network (stDNN) model to uncover latent functional brain dynamics that distinguish male and female brains. Our stDNN model accurately differentiated male and female brains, demonstrating consistently high cross-validation accuracy (>90%), replicability, and generalizability across multisession data from the same individuals and three independent cohorts (N ~ 1,500 young adults aged 20 to 35). Explainable AI (XAI) analysis revealed that brain features associated with the default mode network, striatum, and limbic network consistently exhibited significant sex differences (effect sizes > 1.5) across sessions and independent cohorts. Furthermore, XAI-derived brain features accurately predicted sex-specific cognitive profiles, a finding that was also independently replicated. Our results demonstrate that sex differences in functional brain dynamics are not only highly replicable and generalizable but also behaviorally relevant, challenging the notion of a continuum in male-female brain organization. Our findings underscore the crucial role of sex as a biological determinant in human brain organization, have significant implications for developing personalized sex-specific biomarkers in psychiatric and neurological disorders, and provide innovative AI-based computational tools for future research.

Nuclear Magnetic Resonance Chemical Shift As Highly Explainable Chemical Structure Fingerprints for Anion Exchange Membrane Polymers

Recent shift towards clean energy increased the demand for both fuel cells (used as clean power generator) and water electrolyzer (used as hydrogen supply) significantly. Anion exchange membrane (AEM) serves as a core component for these devices, though its low anion conductivity and durability inhibit their potential for commercialization. Many research and development (R&D) have been done seeking for improvements in AEM1, but current empirical-centric method consumes significant amount of resources, such as cost, labor, and time. To reduce resource consumptions, implementing materials informatics (MI) that allows high-speed screening of materials through a pre-trained AEM polymer machine learning (ML) model is important. However, AEMs are made up of polymers whose chemical structures are complex and hard to represent in a machine-understandable form. Fingerprints are often used to represent chemical structures in numerical forms generated through algorithm2. Majority of these fingerprints are designed for small molecules, not polymers, but they are usually unintuitive and difficult to understand due to their topological nature. In contrast, nuclear magnetic resonance (NMR) chemical shift have long been used in chemistry to identify the chemical structure of a particular sample3. In this study, we aim to utilize the high-resolution nature of NMR chemical shift to identify structural formula as chemical structure fingerprint for ML model, such that a polymer-suited and highly explainable fingerprint can be developed.First, an AEM database containing structural and experimental condition information was built using data extracted from 62 papers. Experimental conditions included were anion conductivity measuring temperature, alkaline stability test measuring condition (test temperature, length of test, and concentration of alkaline solution). Then, the 13C NMR chemical shift for the chemical structure contained in structural information was calculated using ChemDraw. The obtained chemical shifts were converted to numerical strings and is named as “NMR fingerprint”. A new AEM database containing both structural (molar ratio of each building blocks and NMR fingerprints) and experimental condition was used as the training database for ML models. Target variable was set to anion conductivity, and the rest were explanatory variables. ML model used was XGBoost. Cross-validation was used to evaluate the capability of ML models to predict anion conductivity of novel AEM polymers. Prediction logic was analyzed using Shapley additive explanations (SHAP) value.The database built contains data from 62 AEM papers, with 2,197 anion conductivity data points. Each AEM chemical structures present in the database was converted to NMR fingerprints using NMR chemical shifts, obtaining around 2,000 NMR fingerprints for each AEM polymer unit. Together with the experimental conditions and structural information included in the database, the data were used as train-validation dataset for XGBoost. The coefficient of determination (R2) obtained for cross-validated model was 0.9235, implying that the model learnt and determined the relationship between anion conductivity and AEM polymer structure with high accuracy, with the aid of experimental conditions. Then, the prediction logic of the ML model was explored using SHAP values, which are values computed from coalitional game theory, and is used to increase transparency and interpretability of ML models. Analyzing the plot of SHAP values for top 20 important variables used in XGBoost showed that measuring temperature for anion conductivity ranked highest, which is in coherence to the well-known behavior of AEM polymers. Besides, non-experimental condition variables such as 29.8_A ranked into the top 3 important variables. 29.8_A is the chemical shift for alkyl groups attached in between two imidazolium group of AEM polymer, suggesting that the presence or absence of more than one imidazolium group per side chain is important to determine the anion conductivity of an AEM polymer. SHAP values for 29.8_A show that higher feature value (pink color) gives higher impact (positive region of x-axis) to the target variable, inferring that having alkyl groups between imidazolium groups give beneficial effect to anion conductivity. Such ability to explain the prediction logic of ML model shows that using NMR chemical shifts as fingerprints for AEM polymer structures provide intuitive, human-understandable ML prediction logic explanation. Together with the high cross-validation accuracy, NMR chemical shifts hold the potential to not only be a gold standard in expressing polymer structures in machine-understandable form, but also to strongly push the adoption of ML in the AEM polymer field, creating a paradigm shift for AEM R&D.Reference S. Gottesfeld et al., J. Power Sources 2018, 375, 170-184.J. Bajorath, J. Chem. Inf. Comput. Sci. 2001, 41, 2, 233–245.P. Jezzard et al., Adv. Mater. 1992, 4, 2, 82-90. Figure 1

ECG-Based Driving Fatigue Detection Using Heart Rate Variability Analysis with Mutual Information

One of the WHO’s strategies to reduce road traffic injuries and fatalities is to enhance vehicle safety. Driving fatigue detection can be used to increase vehicle safety. Our previous study developed an ECG-based driving fatigue detection framework with AdaBoost, producing a high cross-validated accuracy of 98.82% and a testing accuracy of 81.82%; however, the study did not consider the driver’s cognitive state related to fatigue and redundant features in the classification model. In this paper, we propose developments in the feature extraction and feature selection phases in the driving fatigue detection framework. For feature extraction, we employ heart rate fragmentation to extract non-linear features to analyze the driver’s cognitive status. These features are combined with features obtained from heart rate variability analysis in the time, frequency, and non-linear domains. In feature selection, we employ mutual information to filter redundant features. To find the number of selected features with the best model performance, we carried out 28 combination experiments consisting of 7 possible selected features out of 58 features and 4 ensemble learnings. The results of the experiments show that the random forest algorithm with 44 selected features produced the best model performance testing accuracy of 95.45%, with cross-validated accuracy of 98.65%.

Paper 45: Which Osteochondritis Dissecans Subjects Will Heal Non-Operatively? An Application of Machine Learning Methods to the ROCK Cohort

Paper 45: Which Osteochondritis Dissecans Subjects Will Heal Non-Operatively? An Application of Machine Learning Methods to the ROCK Cohort

Authenticating Tibetan pork in China by tracing the species and geographical features based on stable isotopic and multi-elemental fingerprints

A potential tool for authenticating Tibetan pork based on species and geographical origins was investigated by isotopic and multi-elemental fingerprints to compare Tibetan pork and plain pork from various regions in China. In this study, carbon, nitrogen, hydrogen and oxygen isotope compositions and concentrations of 50 elements of a total of 93 pork samples (Tibetan pork from Tibet and Sichuan, Jilin pork from Jilin and DLY pork from Liaoning) were determined and subjected to multivariate analysis, including linear discriminant analysis (LDA), partial least squares discriminant analysis (PLS-DA) and correlation analysis. The trace elements Fe, Cu, Rb, Mo, Cs, Ir, Co, Sr, Sm and four stable isotopes were determined as potential markers for identifying Tibetan pork. The discriminant model revealed good discrimination among pork samples based on species and regions, and the high accuracies of K-fold cross-validation (95.74% and 97.84%) indicated the reliability and stability of classification models. This study has demonstrated the efficiency of isotopic and multi-elemental technology in identifying Chinese high-altitude indigenous pigs. • Tibetan pork from different regions was discriminated compared to plain pork. • Multi-elements and stable isotopes combined with chemometrics as effective tools. • High cross-validation accuracy for tracing species and origins. • Fe, Cu and δ 13 C are indicators for discriminating high-altitude pork samples.

Developing and validating a highly sensitive platelet clump detection model for the Sysmex haematology analyser.

Mainstream haematology analysers (HAs) are reported to have low detection sensitivity for platelet clumps. In this study, a deep learning (DL) algorithm, convolutional neural network (CNN), was implemented to detect platelet clumps. Adenosine diphosphate (ADP) was used to induce platelet aggregation to mimic platelet clumps detected (PCD) samples. Six types of leukocyte scattergrams were collected from the Sysmex XN-10. Then, multiple CNNs were trained and validated by scattergrams in a fivefold cross-validation (CV) method. Finally, the CNN model with the best CV accuracy was tested with practical routine work samples. A total of 386 samples (190 PCD and 196 negative samples) and 4253 samples (150 PCD and 4103 negative samples) were eligible for CNN training and practical test, respectively. The CNN with the highest CV accuracy was trained by using scattergrams of side scatter (SSC) vs. forward scatter (FSC) from the white count and nucleated red blood cells (WNR) channel, whose mean area under the curve (AUC), accuracy, specificity and sensitivity were 0.968, 0.940, 0.937 and 0.942, respectively, in the CV. In the practical test, the AUC, accuracy, specificity and sensitivity of the CNN were 0.916, 0.961, 0.860 and 0.965, respectively. The dispersed spots presenting around the leucocytes in the WNR channel may be a sign of platelet clumping. This study demonstrates that the CNN algorithms can identify platelet clumps based on optical information from dedicated leukocyte channels and has a higher ability to detect platelet clumps than the XN-10 device's internal algorithm under practical circumstances.

Challenges of deep learning methods for COVID-19 detection using public datasets

Since the COVID-19 pandemic, several research studies have proposed Deep Learning (DL)-based automated COVID-19 detection, reporting high cross-validation accuracy when classifying COVID-19 patients from normal or other common Pneumonia. Although the reported outcomes are very high in most cases, these results were obtained without an independent test set from a separate data source(s). DL models are likely to overfit training data distribution when independent test sets are not utilized or are prone to learn dataset-specific artifacts rather than the actual disease characteristics and underlying pathology. This study aims to assess the promise of such DL methods and datasets by investigating the key challenges and issues by examining the compositions of the available public image datasets and designing different experimental setups. A convolutional neural network-based network, called CVR-Net (COVID-19 Recognition Network), has been proposed for conducting comprehensive experiments to validate our hypothesis. The presented end-to-end CVR-Net is a multi-scale-multi-encoder ensemble model that aggregates the outputs from two different encoders and their different scales to convey the final prediction probability. Three different classification tasks, such as 2-, 3-, 4-classes, are designed where the train–test datasets are from the single, multiple, and independent sources. The obtained binary classification accuracy is 99.8% for a single train–test data source, where the accuracies fall to 98.4% and 88.7% when multiple and independent train–test data sources are utilized. Similar outcomes are noticed in multi-class categorization tasks for single, multiple, and independent data sources, highlighting the challenges in developing DL models with the existing public datasets without an independent test set from a separate dataset. Such a result concludes a requirement for a better-designed dataset for developing DL tools applicable in actual clinical settings. The dataset should have an independent test set; for a single machine or hospital source, have a more balanced set of images for all the prediction classes; and have a balanced dataset from several hospitals and demography. Our source codes and model are publicly available11https://github.com/kamruleee51/CVR-Net. for the research community for further improvements.

Resting-State EEG Signal for Major Depressive Disorder Detection: A Systematic Validation on a Large and Diverse Dataset.

Major depressive disorder (MDD) is a global healthcare issue and one of the leading causes of disability. Machine learning combined with non-invasive electroencephalography (EEG) has recently been shown to have the potential to diagnose MDD. However, most of these studies analyzed small samples of participants recruited from a single source, raising serious concerns about the generalizability of these results in clinical practice. Thus, it has become critical to re-evaluate the efficacy of various common EEG features for MDD detection across large and diverse datasets. To address this issue, we collected resting-state EEG data from 400 participants across four medical centers and tested classification performance of four common EEG features: band power (BP), coherence, Higuchi’s fractal dimension, and Katz’s fractal dimension. Then, a sequential backward selection (SBS) method was used to determine the optimal subset. To overcome the large data variability due to an increased data size and multi-site EEG recordings, we introduced the conformal kernel (CK) transformation to further improve the MDD as compared with the healthy control (HC) classification performance of support vector machine (SVM). The results show that (1) coherence features account for 98% of the optimal feature subset; (2) the CK-SVM outperforms other classifiers such as K-nearest neighbors (K-NN), linear discriminant analysis (LDA), and SVM; (3) the combination of the optimal feature subset and CK-SVM achieves a high five-fold cross-validation accuracy of 91.07% on the training set (140 MDD and 140 HC) and 84.16% on the independent test set (60 MDD and 60 HC). The current results suggest that the coherence-based connectivity is a more reliable feature for achieving high and generalizable MDD detection performance in real-life clinical practice.

Forecasting parasite sharing under climate change.

Species are shifting their distributions in response to climate change. This geographic reshuffling may result in novel co-occurrences among species, which could lead to unseen biotic interactions, including the exchange of parasites between previously isolated hosts. Identifying potential new host-parasite interactions would improve forecasting of disease emergence and inform proactive disease surveillance. However, accurate predictions of future cross-species disease transmission have been hampered by the lack of a generalized approach and data availability. Here, we propose a framework to predict novel host-parasite interactions based on a combination of niche modelling of future host distributions and parasite sharing models. Using the North American ungulates as a proof of concept, we show this approach has high cross-validation accuracy in over 85% of modelled parasites and find that more than 34% of the host-parasite associations forecasted by our models have already been recorded in the literature. We discuss potential sources of uncertainty and bias that may affect our results and similar forecasting approaches, and propose pathways to generate increasingly accurate predictions. Our results indicate that forecasting parasite sharing in response to shifts in host geographic distributions allow for the identification of regions and taxa most susceptible to emergent pathogens under climate change. This article is part of the theme issue 'Infectious disease macroecology: parasite diversity and dynamics across the globe'.

Assessing the use of surrogate species for a more cost-effective traceability of geographic origin using elemental fingerprints of bivalve shells

Illegal, unreported and unregulated (IUU) fishing is a major threat to marine ecosystems and their living resources. IUU fishing promotes fraudulent mislabeling of seafood, which can pose serious risks to consumers’ health if specimens being traded originate from contaminated areas. The development and optimization of reliable tools to confirm the geographic origin of seafood is therefore of utmost importance under these scenarios. These tools are usually based in reference models of a given species that allows to trace the geographic origin of conspecifics. However, to date, the accuracy of such reference models requires the sampling of specimens of the species being surveyed, an issue that may impair the generalized use of this approach by authorities under real-case scenarios. A potential way to overcome this constraint is to apply a model built for a given species (working as surrogate) to trace the geographic origin of another one. The present study evaluated the effect of species-specific and spatial variability associated with the elemental fingerprints (EF) of shells of three commercially important bivalve species (Cerastoderma edule, Ruditapes decussatus and R. philippinarum) from five different locations over the northwestern and western Iberian coast and tested if a reference model based on one of this species could be used to reliably trace the geographic origin of specimens from the two other species. The EF of shells of the three species were significantly different within each location, as well as among different locations within each species. The reference models of the three species presented high cross-validation accuracy (ranging from 95.3% to 99.3%). The tests to infer if these models could reliably trace the geographic origin of samples from another species presented an overall accuracy ranging between 50% and 82.3%. Based on these results, and other requirements that a surrogate species must fulfil, it is possible to conclude that the three species hold the potential to be used as surrogates of each other, although C. edule is the one that better fulfills the requirements of a surrogate species. The optimization proposed in this study improves the cost-efficiency of using EF of bivalve shells as a tool to trace and confirm the geographic origin of bivalves, while also allowing to deliver a prompter reply to authorities.

Rapid and Uninvasive Characterization of Bananas by Hyperspectral Imaging with Extreme Gradient Boosting (XGBoost)

Ripe fruit provides essential nutrients for the human body. To fulfill the needs of consumers, the practice of artificial ripening has become more common. Artificial ripening not only degrades the quality of fruit but also impairs health. The potential of hyperspectral imaging coupled with machine learning to quickly and uninvasively identify differently ripened bananas was explored in this study. A total of 300 banana samples that were naturally ripened or ripened by ethephon or calcium carbide were characterized by their hyperspectral images. To improve the accuracy of classification models and to reduce the effects of noise and irregular surfaces, different preprocessing strategies were investigated. Recursive feature elimination (RFE) and the successive projections algorithm (SPA) were employed to select feature wavelengths. Four classification methods – extreme gradient boosting (XGBoost), support vector machine (SVM), multi-layer perceptron (MLP), and partial least square discriminant analysis (PLS-DA) – were applied for ripening identification. The results showed that the best classification model was XGBoost based on full wavelengths, which achieved an area under the macro-average receiver operating characteristic curve of 0.9796 and high training, cross-validation, and testing accuracy values of 95.02%, 92.31%, and 91.16%, respectively. The RFE-XGBoost model reduces the data dimension and maintains satisfactory performance. Both XGBoost models achieved 100% correct differentiation between naturally ripened and artificially ripened bananas. Hence, the method may be employed to quickly and uninvasively classify differently ripened bananas.

SARS-CoV-2 infection and acute ischemic stroke in Lombardy, Italy

ObjectiveTo characterize patients with acute ischemic stroke related to SARS-CoV-2 infection and assess the classification performance of clinical and laboratory parameters in predicting in-hospital outcome of these patients.MethodsIn the setting of the STROKOVID study including patients with acute ischemic stroke consecutively admitted to the ten hub hospitals in Lombardy, Italy, between March 8 and April 30, 2020, we compared clinical features of patients with confirmed infection and non-infected patients by logistic regression models and survival analysis. Then, we trained and tested a random forest (RF) binary classifier for the prediction of in-hospital death among patients with COVID-19.ResultsAmong 1013 patients, 160 (15.8%) had SARS-CoV-2 infection. Male sex (OR 1.53; 95% CI 1.06–2.27) and atrial fibrillation (OR 1.60; 95% CI 1.05–2.43) were independently associated with COVID-19 status. Patients with COVID-19 had increased stroke severity at admission [median NIHSS score, 9 (25th to75th percentile, 13) vs 6 (25th to75th percentile, 9)] and increased risk of in-hospital death (38.1% deaths vs 7.2%; HR 3.30; 95% CI 2.17–5.02). The RF model based on six clinical and laboratory parameters exhibited high cross-validated classification accuracy (0.86) and precision (0.87), good recall (0.72) and F1-score (0.79) in predicting in-hospital death.ConclusionsIschemic strokes in COVID-19 patients have distinctive risk factor profile and etiology, increased clinical severity and higher in-hospital mortality rate compared to non-COVID-19 patients. A simple model based on clinical and routine laboratory parameters may be useful in identifying ischemic stroke patients with SARS-CoV-2 infection who are unlikely to survive the acute phase.

Prediction of military combat clothing size using decision trees and 3D body scan data

AimTo determine how well decision tree models can predict tailor-assigned uniform sizes using anthropometry data from the New Zealand Defence Force Anthropometry Survey (NZDFAS). This information may inform automatic sizing systems for military personnel. MethodsAnthropometric data from two separate samples of the New Zealand Defence Force military were used. Data on Army personnel from the NZDFAS (n = 583) were used to develop a series of shirt- and trouser-size prediction models based on decision trees. Different combinations of physical, automatic, and post-processed measurements (the latter two derived from a 3D body scan) were trialled, and the models with the highest cross-validation accuracy were retained. The accuracy of these models were then tested on an independent sample of Army recruits (n = 154). ResultsThe automated measurement method (measurements derived automatically by the body scanner software) were the best predictors of shirt size (58.1% accuracy) and trouser size (61.7%), with body weight and waist girth being the strongest predictors. Clothing sizes that were incorrectly predicted by the model where generally one size above or below the tailor-predicted size. ConclusionsAnthropometry measurements, when used with decision tree models, show promise for classifying clothing size. Methodological changes such as fitting gender-specific models, using additional anthropometry variables, and testing other data mining techniques are avenues for future work. More research is required before fully automated body scanning is a viable option for obtaining fast and accurate clothing sizes for military clothing and logistics departments.

Prediction of a Cell-Class-Specific Mouse Mesoconnectome Using Gene Expression Data

Reconstructing brain connectivity at sufficient resolution for computational models designed to study the biophysical mechanisms underlying cognitive processes is extremely challenging. For such a purpose, a mesoconnectome that includes laminar and cell-class specificity would be a major step forward. We analyzed the ability of gene expression patterns to predict cell-class and layer-specific projection patterns and assessed the functional annotations of the most predictive groups of genes. To achieve our goal we used publicly available volumetric gene expression and connectivity data and we trained computational models to learn and predict cell-class and layer-specific axonal projections using gene expression data. Predictions were done in two ways, namely predicting projection strengths using the expression of individual genes and using the co-expression of genes organized in spatial modules, as well as predicting binary forms of projection. For predicting the strength of projections, we found that ridge (L2-regularized) regression had the highest cross-validated accuracy with a median r2 score of 0.54 which corresponded for binarized predictions to a median area under the ROC value of 0.89. Next, we identified 200 spatial gene modules using a dictionary learning and sparse coding approach. We found that these modules yielded predictions of comparable accuracy, with a median r2 score of 0.51. Finally, a gene ontology enrichment analysis of the most predictive gene groups resulted in significant annotations related to postsynaptic function. Taken together, we have demonstrated a prediction workflow that can be used to perform multimodal data integration to improve the accuracy of the predicted mesoconnectome and support other neuroscience use cases.

Non-motor Clinical and Biomarker Predictors Enable High Cross-Validated Accuracy Detection of Early PD but Lesser Cross-Validated Accuracy Detection of Scans Without Evidence of Dopaminergic Deficit.

Background: Early stage (preclinical) detection of Parkinson's disease (PD) remains challenged yet is crucial to both differentiate it from other disorders and facilitate timely administration of neuroprotective treatment as it becomes available.Objective: In a cross-validation paradigm, this work focused on two binary predictive probability analyses: classification of early PD vs. controls and classification of early PD vs. SWEDD (scans without evidence of dopamine deficit). It was hypothesized that five distinct model types using combined non-motor and biomarker features would distinguish early PD from controls with > 80% cross-validated (CV) accuracy, but that the diverse nature of the SWEDD category would reduce early PD vs. SWEDD CV classification accuracy and alter model-based feature selection.Methods: Cross-sectional, baseline data was acquired from the Parkinson's Progressive Markers Initiative (PPMI). Logistic regression, general additive (GAM), decision tree, random forest and XGBoost models were fitted using non-motor clinical and biomarker features. Randomized train and test data partitions were created. Model classification CV performance was compared using the area under the curve (AUC), sensitivity, specificity and the Kappa statistic.Results: All five models achieved >0.80 AUC CV accuracy to distinguish early PD from controls. The GAM (CV AUC 0.928, sensitivity 0.898, specificity 0.897) and XGBoost (CV AUC 0.923, sensitivity 0.875, specificity 0.897) models were the top classifiers. Performance across all models was consistently lower in the early PD/SWEDD analyses, where the highest performing models were XGBoost (CV AUC 0.863, sensitivity 0.905, specificity 0.748) and random forest (CV AUC 0.822, sensitivity 0.809, specificity 0.721). XGBoost detection of non-PD SWEDD matched 1–2 years curated diagnoses in 81.25% (13/16) cases. In both early PD/control and early PD/SWEDD analyses, and across all models, hyposmia was the single most important feature to classification; rapid eye movement behavior disorder (questionnaire) was the next most commonly high ranked feature. Alpha-synuclein was a feature of import to early PD/control but not early PD/SWEDD classification and the Epworth Sleepiness scale was antithetically important to the latter but not former.Interpretation: Non-motor clinical and biomarker variables enable high CV discrimination of early PD vs. controls but are less effective discriminating early PD from SWEDD.

Classification of cardiac arrhythmia using machine learning techniques

Cardiovascular disease is one of the leading causes of death worldwide. Currently, there is an increase in the percentage of people with various heart rhythm disorders. There are permanent (chronic), persistent and paroxysmal form of atrial fibrillation, and the most severe violation is the permanent form. Since the reasons for the development of a certain form of atrial fibrillation are not completely clear, this article presents an analysis of various characteristics that affect the formation of arrhythmias of these species. The most significant signs that can potentially be predictors of different forms of the disease have been identified. Four machine learning methods were used for the analysis: classification trees, logistic regression, random forest, and gradient boosting. The highest cross-validation accuracy was obtained using logistic regression.