Find research datasets worth reusing

FIGURE 2 in Numerical taxonomy and genus-species identification of Czekanowskiales in China based on machine learning

FIGURE 2. Fossils and schematic diagram of the leaf morphology and epidermal structure of Czekanowskia les. (A) Phoenicopsis angustifolia; (B) Czekanowskia rigida; (C) Phoenicopsis speciosa; (D) schematic diagram of the leaf morphology of Czekanowskiales, dotted line ab indicating the leaf cluster length (or lobe length), dotted line cd indicating the leaf cluster width, line ef indicating the lobe width, and angle cbd indicating the angle between bc and bd (the most lateral lobes); (E) schematic diagram of the epidermal structure of Czekanowskiales. Scale bars (A-C) equal 1 cm.

Data from: Evaluating Window Size Effects on Univariate Time Series Forecasting with Machine Learning

In the realm of time series prediction modeling, the window size (w) is a critical hyperparameter that determines the number of time units included in each example provided to a learning model. This hyperparameter is crucial because it allows the learning model to recognize both long-term and short-term trends, as well as seasonal patterns, while reducing sensitivity to random noise. This study aims to elucidate the impact of window size on the performance of machine learning algorithms in univariate time series forecasting tasks. To achieve this, we employed 40 time series from two different domains, conducting experiments with varying window sizes using four types of machine learning algorithms: Bagging, Boosting, Stacking, and a Recurrent Neural Network (RNN) architecture. The results reveal that increasing the window size generally enhances the evaluation metric values up to a stabilization point, beyond which further increases do not significantly improve predictive accuracy. This stabilization effect was observed in both domains when w values exceeded 100 time steps. Moreover, the study found that RNN architectures do not consistently outperform ensemble models in various univariate time series forecasting scenarios.

opencc-zeroJul 2024View details →

dryad40/100

Data from: Performance of unmarked abundance models with data from machine-learning classification of passive acoustic recordings

The ability to conduct cost-effective wildlife monitoring at scale is rapidly increasing due to availability of inexpensive autonomous recording units (ARUs) and automated species recognition, presenting a variety of advantages over human-based surveys. However, estimating abundance with such data collection techniques remains challenging because most abundance models require data that are difficult for low-cost monoaural ARUs to gather (e.g., counts of individuals, distance to individuals), especially when using the output of automated species recognition. Statistical models that do not require counting or measuring distances to target individuals in combination with low-cost ARUs provide a promising way of obtaining abundance estimates for large-scale wildlife monitoring projects but remain untested. We present a case study using avian field data collected in forests of Pennsylvania during the Spring of 2020 and 2021 using both traditional point counts and passive acoustic monitoring at the same locations. We tested the ability of the Royle-Nichols and time-to-detection models to estimate abundance of two species from detection histories generated by applying a machine-learning classifier to ARU-gathered data. We compared abundance estimates from these models to estimates from the same models fit using point-count data and to two additional models appropriate for point counts, the N-mixture model and distance models. We found that the Royle-Nichols and time-to-detection models can be used with ARU data to produce abundance estimates similar to those generated by a point-count based study but with greater precision. ARU-based models produced confidence or credible intervals that were on average 31.9% ( 11.9 SE) smaller than their point-count counterpart. Our findings were consistent across two species with differing relative abundance and habitat use patterns. The higher precision of models fit using ARU data is likely due to higher cumulative detection probability, which itself may be the result of greater survey effort using ARUs and machine-learning classifiers to sample significantly more time for focal species at any given point. Our results provide preliminary support the use of ARUs in abundance-based study applications, and thus may afford researchers a better understanding of habitat quality and population trends, while allowing them to make more informed conservation actions and recommendations.

opencc-by-4.0Jun 2024View details →

Dataset for "Machine Learning Driven Sensitivity Analysis of E3SM Land Model Parameters for Wetland Methane Emissions"

This dataset is a part of the paper "Machine Learning Driven Sensitivity Analysis of E3SM Land Model Parameters for Wetland Methane Emissions", accepted for publication in the Journal of Advances in Modeling Earth Systems (JAMES). <h2>Contents</h2> This dataset includes: <ul> <li>lhs-gen-190.csv: Training input LHS samples generated by <code>lhsgen.py</code>.</li> <li>lhs-gen-50-test.csv: Test input LHS samples generated by <code>lhsgen.py</code>.</li> <li>190-elm-samples.csv: Training input perturbed parameter samples for performing ELM simulations.</li> <li>50-elm-test-samples.csv: Test input perturbed parameter samples for performing ELM simulations.</li> <li>train_CH-CHA.csv: Contains the five ELM simulation output flux values for 240 samples (190 train + 50 test).</li> <li>lhsgen.py: Script for generating Latin Hypercube Samples.</li> <li>gpr-fit-new.py: Script for fitting Gaussian Process Regression (GPR) models.</li> <li>sobol-new.py: Script for performing Sobol sensitivity analysis.</li> </ul> <h2>Usage</h2> <ol> <li>lhsgen.py: <ul> <li>Use this script to generate the Latin Hypercube Samples for parameter sampling.</li> </ul> </li> <li>gpr-fit-new.py: <ul> <li>This script fits GPR models using the training samples provided in <code>lhs-gen-190.csv</code>.</li> <li>It tests the models using the input testing samples in <code>lhs-gen-50-test.csv</code>.</li> <li>The fitted GPR models are stored as <code>.joblib</code> files in the <code>gpr_models</code> directory.</li> <li>Corresponding cross-validation and R-squared values are stored in <code>.xlsx</code> files.</li> </ul> </li> <li>sobol-new.py: <ul> <li>This script performs Sobol sensitivity analysis using the fitted GPR models by reading the .joblib files.</li> <li>The Sobol indices are written to <code>.xlsx</code> files in the <code>results</code> directory.</li> </ul> </li> </ol>

Cellpose training data and scripts from "Machine learning for histological annotation and quantification of cortical layers"

This Workflow contains all the material necessary to reproduce the cells detection, thanks to the QuPath performed in the paper  "Machine learning for histological annotation and quantification of cortical layers" Inside this workflow and dataset, you will find the following folders <ol> <li>QuPath Training Project: A QuPath 0.5.0 project containing all the manual annotations (ground truths) used to train the cellpose model, as well as the script to start the training</li> <li>Training Images and Demo Images: The raw whole slide scanner images needed by the above QuPath project</li> <li>Model: The fodler containing the trained cellpose model</li> <li>cellpose-training Folder: The exported raw and ground truth images that the above cellpose model was trained on</li> <li>Scripts: The QuPath scripts, also located in their respective QuPath projects, that were created for this whole workflow</li> <li>QC: A Jupyter notebook, based on ZeroCostDL4Mic that computes quality metrics in order to assess the performance of the trained cellpose model. The folder also contains the resulting metrics.</li> </ol> Installation and Use If you are going to use the QuPath projects, you need a local QuPath Installation https://qupath.github.io/ that is configured to run the QuPath Cellpose Extension https://github.com/BIOP/qupath-extension-cellpose as well as a working Cellpose installation https://github.com/MouseLand/cellpose Instructions for installation are available from the links above. After that, you should be able to open the QuPath project, navigate to the "Automate > Project scripts" menu and locate the script you wish to run. 1. train a cell segmentation algorithm in the context of the rat brain Layer  Boundaries project  2. trigger cell segmentation from a QuPath project in a semi-automated pipeline

Shaping History: Advanced Machine Learning Techniques for the Analysis and Dating of Cuneiform Tablets over Three Millennia - Datasets

Our research leverages advanced deep learning methods to classify cuneiform tablets by their historical periods, focusing on shape analysis rather than textual content. Utilizing a dataset of over 94,000 images from the Cuneiform Digital Library Initiative, we introduce a novel toolset powered by Variational Auto-Encoders (VAEs) to enhance model interpretability. By highlighting the predictive power of tablet silhouettes for historical period classification with a ResNet model reaching 61\% macro-F1 score,, our approach allows researchers to explore changes in tablet shapes across different eras. This methodology not only complements traditional archaeological methods but also enriches the field of document analysis and diplomatics, offering valuable tools for historians and epigraphists to understand ancient Mesopotamian cultures. The Datasets attached include: <ul> <li>All CDLI IDs used and their URLs on CDLI</li> <li>All tablet silhouettes - Black and white representations</li> <li>A height-to-width ratio table, of the largest component extracted from the silhouettes.</li> <li>VAE encodings per tablet, extracted from the VAE model</li> </ul>

Brazilian Cohort for Predicting Cardiovascular Events Using Machine Learning (PRE-CARE ML project)

The project PRE-CARE ML addresses the development and internal and external validation of predictive models for the assessment of risks of major adverse cardiovascular events.  Global and local interpretability analyses of predictions were conducted towards improving model reliability and tailoring preventive interventions.  The models were trained and validated in a retrospective cohort with the use of data from Hospital das Clínicas da Faculdade de Medicina de Ribeirao Preto, Brazil The International Classification of Diseases (ICD-10) classified patients with MACE (case group). (see Table I for the ICD-10 codes that defined MACE). Table 1: ICD-10 Codees for MACE definition <table> <tbody> <tr> <td>ICD-10</td> <td>Description</td> </tr> <tr> <td>I20</td> <td>Angina pectoris</td> </tr> <tr> <td>I21</td> <td>Acute myocardial infarction</td> </tr> <tr> <td>I24</td> <td>Other acute ischaemic heart diseases</td> </tr> <tr> <td>I46</td> <td>Cardiac arrest</td> </tr> <tr> <td>I63</td> <td>Cerebral infarction</td> </tr> <tr> <td>I64</td> <td>Stroke, not specified as hemorrhage or infarction</td> </tr> <tr> <td>I71</td> <td>Aortic aneurysm and dissection</td> </tr> <tr> <td>I74</td> <td>Arterial embolism and thrombosis</td> </tr> </tbody> </table> Only a patient’s first MACE was considered, and all previous hospitalizations within a 5-year window were considered MACE. The control group (non-MACE) involved hospitalizations with no MACE and no death within a 5-year window (between 2017 and 2022). A sample of 6,000 MACE (labeled as 1) cases and 12,000 non-MACE (labeled as 0) ones was constructed with data from HCFMRP for training and internal validation purposes. Another balanced MIMIC IV sample of 8,000 MACE cases and 8,000 nonMACE cases was used for external validation.

opencc-by-4.0Jul 2022View details →

Bio-ML: Machine Learning-Friendly Biomedical Datasets for Equivalence and Subsumption Ontology Matching

<blockquote> This version is used in the Bio-ML track of the OAEI 2024; the only change compared to the OAEI 2023 is the deletion of certain training subsumption mappings. </blockquote>   <h3>Overview</h3> The purpose of these datasets is to support equivalence and subsumption ontology matching. There are five ontology pairs extracted from MONDO and UMLS: <table> <tbody> <tr> <td>Source</td> <td>Task</td> <td>Category</td> <td>#SrcCls</td> <td>#TgtCls</td> <td>#Ref (equiv)</td> <td>#Ref (subs)</td> </tr> <tr> <td>Mondo</td> <td>OMIM-ORDO</td> <td>Disease</td> <td>9,648</td> <td>9,275</td> <td>3,721</td> <td>103</td> </tr> <tr> <td>Mondo</td> <td>NCIT-DOID</td> <td>Disease</td> <td>15,762</td> <td>8,465</td> <td>4,686</td> <td>3,338 (-1)</td> </tr> <tr> <td>UMLS</td> <td>SNOMED-FMA</td> <td>Body</td> <td>34,418</td> <td>88,955</td> <td>7,256</td> <td>5,453 (-53)</td> </tr> <tr> <td>UMLS</td> <td>SNOMED-NCIT</td> <td>Pharm</td> <td>29,500</td> <td>22,136</td> <td>5,803</td> <td>4,224 (-1)</td> </tr> <tr> <td>UMLS</td> <td>SNOMED-NCIT</td> <td>Neoplas</td> <td>22,971</td> <td>20,247</td> <td>3,804</td> <td>213</td> </tr> </tbody> </table> The "-" numbers reflect the changes due to lthe deletion of certain training subsumption mappings. The main track is available at "bio-ml", where each pair is associated with a task folder, containing the source and target ontologies, reference equivalence mappings (in "refs_equiv"), reference subsumption mappings ("refs_subs").  The special sub-track is available at "bio-llm", where each pair is associated with a task folder, containing the source and target ontologies, and the test candidate mappings.    <h3>Citation</h3> Bio-ML (Main Track) <pre>``` @inproceedings{he2022machine, title={Machine learning-friendly biomedical datasets for equivalence and subsumption ontology matching}, author={He, Yuan and Chen, Jiaoyan and Dong, Hang and Jim{\'e}nez-Ruiz, Ernesto and Hadian, Ali and Horrocks, Ian}, booktitle={International Semantic Web Conference}, pages={575--591}, year={2022}, organization={Springer} } ```</pre> Bio-LLM (Sub-track) <pre>``` @article{he2023exploring, title={Exploring large language models for ontology alignment}, author={He, Yuan and Chen, Jiaoyan and Dong, Hang and Horrocks, Ian}, journal={arXiv preprint arXiv:2309.07172}, year={2023} } ```</pre>   <h3>Important Links</h3> <ul> <li>See detailed documentation at: <a href="https://krr-oxford.github.io/DeepOnto/bio-ml">https://krr-oxford.github.io/DeepOnto/bio-ml</a>.</li> <li>See the OAEI Bio-ML track at: <a href="https://www.cs.ox.ac.uk/isg/projects/ConCur/oaei/">https://www.cs.ox.ac.uk/isg/projects/ConCur/oaei/</a></li> <li>See our resource paper for the original Bio-ML at <a href="https://arxiv.org/abs/2205.03447">arxiv</a> or <a href="https://link.springer.com/chapter/10.1007/978-3-031-19433-7_33">springer</a> (accepted at ISWC-2022 and nominated as the best resource paper candidate). See our poster paper for the Bio-LLM sub-track at <a href="https://arxiv.org/abs/2309.07172">arxiv </a>(accepted at ISWC-2023 Posters & Demos).</li> </ul>   <h3>Changelog</h3> The only change in this version compared to the OAEI 2023 is the deletion of certain training subsumption mappings that can be directly exploited through deductive reasoning.

opencc-by-4.0Aug 2024View details →

Radiative-transfer dataset for "Distilling machine learning's added value: Pareto fronts in atmospheric applications"

This dataset goes with the journal paper "Distilling machine learning's added value: Pareto fronts in atmospheric applications" by T. Beucler, A. Grundner, S. Shamekh, P. Ukkonen, M. Chantry, and R. Lagerquist. Subdirectory "training" contains unnormalized (in physical units) training data.  Subdirectories "validation" and "testing" contain unnormalized validation and testing data.  Subdirectory "training/for_pareto_paper_2024/simple" contains training data from the simple (clear-sky) dataset discussed in the paper; subdirectory "training/for_pareto_paper_2024/complex" contains training data from the complex (multi-cloud) dataset discussed in the paper.  Subdirectories "validation/for_pareto_paper_2024/simple" and "validation/for_pareto_paper_2024/complex" are analogous but for the validation data; subdirectories "testing/for_pareto_paper_2024/simple" and "testing/for_pareto_paper_2024/complex" are analogous but for the testing data. Subdirectories beginning with "normalized_predictors" -- "normalized_predictors/training", "normalized_predictors/validation", "normalized_predictors/testing", "normalized_predictors/training/for_pareto_paper_2024/simple", "normalized_predictors/training/for_pareto_paper_2024/complex", etc. -- are analogous to the above but containing normalized predictors (in z-scores rather than physical units). Every file -- after unzipping, so that the extension is ".nc" rather than ".nc.gz" -- can be read by `example_io.read_file` in the ml4rt library (https://github.com/thunderhoser/ml4rt).

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure7. Model 3D of women body.

In two cases, using the SVM classifier and Random Forest with trees 100, 200, 300, 400, 500 datasets before and after optimization commented as follows: the running time of Random Forest is greater comparing with SVM, because more trees are generated, many cases will be considered. In particular, increasing the number of trees, while labeling is long, but Random Forest provides higher accuracy SVM. Based on anthropometric features and machine learning algorithms, we have built an Android app in the smartphone environment. This app can automatically extrac tanthropometric features (12 features). The user must stand in front of the smartphone camera and takes 2 pictures. Then input their height (centimeters) for calibration. The application automatically extracts human parameters to enable adequate 3D models reconstruction. The results of the Android application are demonstrated in figure 7 .

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure 3. (3.a) – The flowchart of Graph cuts method; (3.b)- the result of Graph cuts image segmentation.

Figure 3 describes the steps implemented Graph cuts algorithm for the segmentation of human body parts. The results obtained are 5 main sections that include the hands, the legs, the center of the body (chest, waist, hips), and the head. The result of the display image is taken from the human image database, which was collected by us (Нгуен, 2016).

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure 2. Human body sizes for men/women.

We propose an efficient, simple and robust human body feature extraction based on the front and side images of a human body. Description of anthropometric data - men/women: Dataset based on an experiment is used to test the system data describing the anthropometric features of men, includes 12 sizes of the human body, which are presented in figure 2.

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure 6. The result of building a 3D model based on RF and SVM classification with "Important features".

From the chart of figure 6, we found that "Important Features" gave the best 3D model, which fits with the object in the image. The pattern is close to 90% compared with the true size. Apply classification algorithm RF increases the accuracy of the results and reduces computing time for the program. There are many methods for data classifying. One of them is the method of the support vector machine (SVM). The SVM method is represented by Vladimir N. Vapnik (1995) in Support Vector Machines (SVM) - a set of learning algorithms similar with the supervisor has two main tasks: the classification and the regression analysis. In this article we use the method of the SVM classification problem for the size of the human body with 5 classes to compare the performance between SVM methods and Random Forest algorithm.

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure 1. Flowchart of anthropometric system

Our purpose is to develop an automatic measurement and modeling system based on 2D images (front and side images). This system used to image processing methods and machine learning algorithms. Our system has 3 main parts; there are human body feature extraction, training and testing processes, and the classification for new data. The novelty of our approach: - Classification of anthropometric features based on machine learning algorithms. - Development a non-contact anthropometric program for the smartphones on operation system Android. - Construction of a 3D-model of the human body based on the results of anthropometric features extraction. Our system can also be used to integrate to different environments, such as online shopping websites to support users fitting their clothes sizes and medical applications. The flowchart of our anthropometric system is described in figure 1.

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure 5. Flowchart of data classification

The Random Forest is a powerful classification method because of the following. First, errors are minimized as a result of a random forest, synthesizing through training (learner). The second, random choice at every stage in the Random Forest will reduce the correlation between the learners in the synthesis of the results. In addition, we also found that the total error of layered forest trees depends on their individual errors in forest trees, as well as the correlation between the trees. The article uses the wrapper model (Christopher Tong, 2000) with the objective function for the evaluation, Random Forest algorithm is shown in figure 5.

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure 8. Model 3D of man body

BRAIN Journal-Automatic Anthropometric System Development Using Machine Learning-Figure 4. Flowchart and results of ICP algorithm

The key concept of the standard ICP algorithm can be summarized in two steps: - Compute correspondences between the two scans. - Compute a transformation which minimizes the distance between corresponding points. It is forced to add a maximum matching threshold dmax. In most implementations of ICP, the choice of dmax represents a tradeoff between convergence and accuracy. A low-value result in bad convergence, a large value causes incorrect correspondences to pull the final alignment away from the correct value. Figure 4 describes the steps of the algorithm which determines the point features closest to object boundary. The result of the algorithm is described by images cut from the program (Нгуен, 2016)

opencc-by-sa-4.0Mar 2018View details →

Noninvasive detection of macrophage activation with single-cell resolution through machine learning

Data related to the article "Noninvasive detection of macrophage activation with single-cell resolution through machine learning". The package contains 2 folders: - RawData:   This package contains raw data and examples of processing to extract the              variables employed to train and assess the models. - Variables: This package contains the extracted data from the various experiments              showed in the article.

opencc-by-4.0Jul 2018View details →

Partitioned Image Data for Machine Learning Analysis of Molecular Biology Figures

Corpus Composition This data collection provides four types of hand-curated images from open access research articles images. The types are: <ol> <li>chart (n=811): data displays such as bar charts, scatterplots, line graphs, etc.</li> <li>diagram (n=816): any general conceptual diagram</li> <li>gel (n=1182): the output of electrophoresis experiments in Northern, Western, or Southern Blot experiments. </li> <li>histology (n=3458): microscope images of tissue with histological staining</li> </ol> The images are simply organized in subdirectories as individual files. File names are based on PubMed Id and Figure number.