Mixed-type Data Research Articles

Mixed-type data generation method based on generative adversarial networks

Data-driven based deep learing has become a key research direction in the field of artificial intelligence. Abundant training data is a guarantee for building efficient and accurate models. However, due to the privacy protection policy, research institutions are often limited to obtain a large number of training data, which would lead to a lack of training sets circumstance. In this paper, a mixed-type data generation model based on generative adversarial networks is proposed to synthesize fake data that have the same distribution with the real data, so as to supplement the real data and increase the number of available samples. The model first pre-trains the autoencoder which maps given dataset into a low-dimensional continuous space. Then, the generator constructed in the low-dimension space is obtained by training it adversarially with discriminator constructed in the original space. Since the constructed discriminator not only consider the loss of the continuous attributes but also the labeled attributes, the generator nets formed by the generator and the decoder can effectively learn the intrinsic distribution of the mixed data. We evaluate the proposed method both in the independent distribution of the attribute and in the relationship of the attributes, and the experiment results show that the proposed generate method has a better performance in preserve the intrinsic distribution compared with other generation algorithms based on deep learning.

A machine learning approach for identification of gastrointestinal predictors for the risk of COVID-19 related hospitalization.

Background and aimCOVID-19 can be presented with various gastrointestinal symptoms. Shortly after the pandemic outbreak, several machine learning algorithms were implemented to assess new diagnostic and therapeutic methods for this disease. The aim of this study is to assess gastrointestinal and liver-related predictive factors for SARS-CoV-2 associated risk of hospitalization.MethodsData collection was based on a questionnaire from the COVID-19 outpatient test center and from the emergency department at the University Hospital in combination with the data from internal hospital information system and from a mobile application used for telemedicine follow-up of patients. For statistical analysis SARS-CoV-2 negative patients were considered as controls in three different SARS-CoV-2 positive patient groups (divided based on severity of the disease). The data were visualized and analyzed in R version 4.0.5. The Chi-squared or Fisher test was applied to test the null hypothesis of independence between the factors followed, where appropriate, by the multiple comparisons with the Benjamini Hochberg adjustment. The null hypothesis of the equality of the population medians of a continuous variable was tested by the Kruskal Wallis test, followed by the Dunn multiple comparisons test. In order to assess predictive power of the gastrointestinal parameters and other measured variables for predicting an outcome of the patient group the Random Forest machine learning algorithm was trained on the data. The predictive ability was quantified by the ROC curve, constructed from the Out-of-Bag data. Matthews correlation coefficient was used as a one-number summary of the quality of binary classification. The importance of the predictors was measured using the Variable Importance. A 2D representation of the data was obtained by means of Principal Component Analysis for mixed type of data. Findings with the p-value below 0.05 were considered statistically significant.ResultsA total of 710 patients were enrolled in the study. The presence of diarrhea and nausea was significantly higher in the emergency department group than in the COVID-19 outpatient test center. Among liver enzymes only aspartate transaminase (AST) has been significantly elevated in the hospitalized group compared to patients discharged home. Based on the Random Forest algorithm, AST has been identified as the most important predictor followed by age or diabetes mellitus. Diarrhea and bloating have also predictive importance, although much lower than AST.ConclusionSARS-CoV-2 positivity is connected with isolated AST elevation and the level is linked with the severity of the disease. Furthermore, using the machine learning Random Forest algorithm, we have identified the elevated AST as the most important predictor for COVID-19 related hospitalizations.

Decomposition of variation of mixed variables by a latent mixed Gaussian copula model.

Many biomedical studies collect data of mixed types of variables from multiple groups of subjects. Some of these studies aim to find the group-specific and the common variation among all these variables. Even though similar problems have been studied by some previous works, their methods mainly rely on the Pearson correlation, which cannot handle mixed data. To address this issue, we propose a latent mixed Gaussian copula (LMGC) model that can quantify the correlations among binary, ordinal, continuous, and truncated variables in a unified framework. We also provide a tool to decompose the variation into the group-specific and the common variation over multiple groups via solving a regularized M-estimation problem. We conduct extensive simulation studies to show the advantage of our proposed method over the Pearson correlation-based methods. We also demonstrate that by jointly solving the M-estimation problem over multiple groups, our method is better than decomposing the variation group by group. We also apply our method to a Chlamydia trachomatis genital tract infection study to demonstrate how it can be used to discover informative biomarkers that differentiatepatients.

Personas for Better Targeted eHealth Technologies: User-Centered Design Approach.

BackgroundThe full potential of eHealth technologies to support self-management and disease management for patients with chronic diseases is not being reached. A possible explanation for these lacking results is that during the development process, insufficient attention is paid to the needs, wishes, and context of the prospective end users. To overcome such issues, the user-centered design practice of creating personas is widely accepted to ensure the fit between a technology and the target group or end users throughout all phases of development.ObjectiveIn this study, we integrate several approaches to persona development into the Persona Approach Twente to attain a more holistic and structured approach that aligns with the iterative process of eHealth development.MethodsIn 3 steps, a secondary analysis was carried out on different parts of the data set using the Partitioning Around Medoids clustering method. First, we used health-related electronic patient record data only. Second, we added person-related data that were gathered through interviews and questionnaires. Third, we added log data.ResultsIn the first step, 2 clusters were found, with average silhouette widths of 0.12 and 0.27. In the second step, again 2 clusters were found, with average silhouette widths of 0.08 and 0.12. In the third step, 3 clusters were identified, with average silhouette widths of 0.09, 0.12, and 0.04.ConclusionsThe Persona Approach Twente is applicable for mixed types of data and allows alignment of this user-centered design method to the iterative approach of eHealth development. A variety of characteristics can be used that stretches beyond (standardized) medical and demographic measurements. Challenges lie in data quality and fitness for (quantitative) clustering.

Differentially private synthetic mixed-type data generation for unsupervised learning

We introduce the DP-auto-GAN framework for synthetic data generation, which combines the low dimensional representation of autoencoders with the flexibility of Generative Adversarial Networks (GANs). This framework can be used to take in raw sensitive data and privately train a model for generating synthetic data that will satisfy similar statistical properties as the original data. This learned model can generate an arbitrary amount of synthetic data, which can then be freely shared due to the post-processing guarantee of differential privacy. Our framework is applicable to unlabeled mixed-type data, that may include binary, categorical, and real-valued data. We implement this framework on both binary data (MIMIC-III) and mixed-type data (ADULT), and compare its performance with existing private algorithms on metrics in unsupervised settings. We also introduce a new quantitative metric able to detect diversity, or lack thereof, of synthetic data.

Using Link-Based Consensus Clustering for Mixed-Type Data Analysis

A mix between numerical and nominal data types commonly presents many modern-age data collections. Examples of these include banking data, sales history and healthcare records, where both continuous attributes like age and nominal ones like blood type are exploited to characterize account details, business transactions or individuals. However, only a few standard clustering techniques and consensus clustering methods are provided to examine such a data thus far. Given this insight, the paper introduces novel extensions of link-based cluster ensemble, and that are accurate for analyzing mixed-type data. They promote diversity within an ensemble through different initializations of the k-prototypes algorithm as base clusterings and then refine the summarized data using a link-based approach. Based on the evaluation metric of NMI (Normalized Mutual Information) that is averaged across different combinations of benchmark datasets and experimental settings, these new models reach the improved level of 0.34, while the best model found in the literature obtains only around the mark of 0.24. Besides, parameter analysis included herein helps to enhance their performance even further, given relations of clustering quality and algorithmic variables specific to the underlying link-based models. Moreover, another significant factor of ensemble size is examined in such a way to justify a tradeoff between complexity and accuracy.

A Novel Attention-Based Multi-Modal Modeling Technique on Mixed Type Data for Improving TFT-LCD Repair Process

In Thin-Film Transistor Liquid-Crystal Display (TFT-LCD) manufacturing, conducting a machine learning based system with multiple data types has become actively desired to solve complicated problems. This paper proposes a multi-modal learning approach: <i>TabVisionNet</i>, which is modeled by utilizing the information from both tabular data and image data. A novel attention mechanism called <i>Sequential Decision Attention</i> was integrated into the multi-modal modeling framework that improves the comprehension of the information from two modalities. This cross-modal attention mechanism can capture the complex relationship between modalities then gain better generalization and faster convergence in the training process. Conducting an experiment, the performance of our novel approach was significantly better than single-modal and other multi-modal learning approaches in our real case scenario.

Hierarchy-based semantic embeddings for single-valued & multi-valued categorical variables

In low-resource domains, it is challenging to achieve good performance using existing machine learning methods due to a lack of training data and mixed data types (numeric and categorical). In particular, categorical variables with high cardinality pose a challenge to machine learning tasks such as classification and regression because training requires sufficiently many data points for the possible values of each variable. Since interpolation is not possible, nothing can be learned for values not seen in the training set. This paper presents a method that uses prior knowledge of the application domain to support machine learning in cases with insufficient data. We propose to address this challenge by using embeddings for categorical variables that are based on an explicit representation of domain knowledge (KR), namely a hierarchy of concepts. Our approach is to 1. define a semantic similarity measure between categories, based on the hierarchy—we propose a purely hierarchy-based measure, but other similarity measures from the literature can be used—and 2. use that similarity measure to define a modified one-hot encoding. We propose two embedding schemes for single-valued and multi-valued categorical data. We perform experiments on three different use cases. We first compare existing similarity approaches with our approach on a word pair similarity use case. This is followed by creating word embeddings using different similarity approaches. A comparison with existing methods such as Google, Word2Vec and GloVe embeddings on several benchmarks shows better performance on concept categorisation tasks when using knowledge-based embeddings. The third use case uses a medical dataset to compare the performance of semantic-based embeddings and standard binary encodings. Significant improvement in performance of the downstream classification tasks is achieved by using semantic information.

Spectral Clustering of Mixed-Type Data

Cluster analysis seeks to assign objects with similar characteristics into groups called clusters so that objects within a group are similar to each other and dissimilar to objects in other groups. Spectral clustering has been shown to perform well in different scenarios on continuous data: it can detect convex and non-convex clusters, and can detect overlapping clusters. However, the constraint on continuous data can be limiting in real applications where data are often of mixed-type, i.e., data that contains both continuous and categorical features. This paper looks at extending spectral clustering to mixed-type data. The new method replaces the Euclidean-based similarity distance used in conventional spectral clustering with different dissimilarity measures for continuous and categorical variables. A global dissimilarity measure is than computed using a weighted sum, and a Gaussian kernel is used to convert the dissimilarity matrix into a similarity matrix. The new method includes an automatic tuning of the variable weight and kernel parameter. The performance of spectral clustering in different scenarios is compared with that of two state-of-the-art mixed-type data clustering methods, k-prototypes and KAMILA, using several simulated and real data sets.

Generalized Factor Model for Ultra-High Dimensional Correlated Variables with Mixed Types

As high-dimensional data measured with mixed-type variables gradually become prevalent, it is particularly appealing to represent those mixed-type high-dimensional data using a much smaller set of so-called factors. Due to the limitation of the existing methods for factor analysis that deal with only continuous variables, in this article, we develop a generalized factor model, a corresponding algorithm and theory for ultra-high dimensional mixed types of variables where both the sample size n and variable dimension p could diverge to infinity. Specifically, to solve the computational problem arising from the non-linearity and mixed types, we develop a two-step algorithm so that each update can be carried out in parallel across variables and samples by using an existing package. Theoretically, we establish the rate of convergence for the estimators of factors and loadings in the presence of nonlinear structure accompanied with mixed-type variables when both n and p diverge to infinity. Moreover, since the correct specification of the number of factors is crucial to both the theoretical and the empirical validity of factor models, we also develop a criterion based on a penalized loss to consistently estimate the number of factors under the framework of a generalized factor model. To demonstrate the advantages of the proposed method over the existing ones, we conducted extensive simulation studies and also applied it to the analysis of the NFBC1966 dataset and a cardiac arrhythmia dataset, resulting in more predictive and interpretable estimators for loadings and factors than the existing factor model.

Structural inference of time‐varying mixed graphical models

With the rapid advancement of biotechnology, there arise new challenges to discover regulatory and co‐action relationships among heterogeneous biological features as well as to capture significant topological changes when data are observed across time. This paper is devoted to joint structural estimation of time‐varying mixed graphical models based on multivariate data over a series of time points. Assuming the graph topology changes gradually over time, we establish a flexible local estimator to fully exploit the structural smoothness. Utilizing variational likelihood inference, we impose a group lasso penalty to integrate information from nearby time points. In order to reduce the algorithmic complexity, we propose an accelerated alternating direction method of multipliers (ADMM)‐based algorithm exploiting the block diagonal structure to adapt our problem for large sparse networks. Practical merits of our method are exhibited through synthetic networks with mixed data types. Ultimately, we illustrate the real application by incorporating multi‐platform data from the PsychENCODE human brain development project and detect the evolution of gene regulatory networks along different stages of human brain development.

Analyzing mixed-type data by using word embedding for handling categorical features

Most of real-world datasets are of mixed type including both numeric and categorical attributes. Unlike numbers, operations on categorical values are limited, and the degree of similarity between distinct values cannot be measured directly. In order to properly analyze mixed-type data, dedicated methods to handle categorical values in the datasets are needed. The limitation of most existing methods is lack of appropriate numeric representations of categorical values. Consequently, some of analysis algorithms cannot be applied. In this paper, we address this deficiency by transforming categorical values to their numeric representation so as to facilitate various analyses of mixed-type data. In particular, the proposed transformation method preserves semantics of categorical values with respect to the other values in the dataset, resulting in better performance on data analyses including classification and clustering. The proposed method is verified and compared with other methods on extensive real-world datasets.

AdRoit is an accurate and robust method to infer complex transcriptome composition

Bulk RNA sequencing provides the opportunity to understand biology at the whole transcriptome level without the prohibitive cost of single cell profiling. Advances in spatial transcriptomics enable to dissect tissue organization and function by genome-wide gene expressions. However, the readout of both technologies is the overall gene expression across potentially many cell types without directly providing the information of cell type constitution. Although several in-silico approaches have been proposed to deconvolute RNA-Seq data composed of multiple cell types, many suffer a deterioration of performance in complex tissues. Here we present AdRoit, an accurate and robust method to infer the cell composition from transcriptome data of mixed cell types. AdRoit uses gene expression profiles obtained from single cell RNA sequencing as a reference. It employs an adaptive learning approach to alleviate the sequencing technique difference between the single cell and the bulk (or spatial) transcriptome data, enhancing cross-platform readout comparability. Our systematic benchmarking and applications, which include deconvoluting complex mixtures that encompass 30 cell types, demonstrate its preferable sensitivity and specificity compared to many existing methods as well as its utilities. In addition, AdRoit is computationally efficient and runs orders of magnitude faster than most methods.

Chrono::GPU: An Open-Source Simulation Package for Granular Dynamics Using the Discrete Element Method

We report on an open-source, publicly available C++ software module called Chrono::GPU, which uses the Discrete Element Method (DEM) to simulate large granular systems on Graphics Processing Unit (GPU) cards. The solver supports the integration of granular material with geometries defined by triangle meshes, as well as co-simulation with the multi-physics simulation engine Chrono. Chrono::GPU adopts a smooth contact formulation and implements various common contact force models, such as the Hertzian model for normal force and the Mindlin friction force model, which takes into account the history of tangential displacement, rolling frictional torques, and cohesion. We report on the code structure and highlight its use of mixed data types for reducing the memory footprint and increasing simulation speed. We discuss several validation tests (wave propagation, rotating drum, direct shear test, crater test) that compare the simulation results against experimental data or results reported in the literature. In another benchmark test, we demonstrate linear scaling with a problem size up to the GPU memory capacity; specifically, for systems with 130 million DEM elements. The simulation infrastructure is demonstrated in conjunction with simulations of the NASA Curiosity rover, which is currently active on Mars.

Retrospective identification of latent subgroups of emergency department patients: A machine learning approach.

This research aims to (i) identify latent subgroups of ED presentations in Australian public EDs using a data-driven approach and (ii) compare clinical, socio-demographic and time-related characteristics of ED presentations broadly using the subgroups. We examined presentations to four public hospital EDs in Queensland from 2009 to 2014. An unsupervised machine learning algorithm, Clustering Large Applications, was used to cluster ED presentations. There were six subgroups common across the EDs, primarily distinguishable by age, and subsequently by triage category, ED length of stay, arrival mode, departure status and several time-related attributes. Around 10% to 30% of the total presentations had high resource utilisation, with half of these from older patients (55+ years). ED resource utilisation per population was highest among the oldest cohort (75+ years). Children and young adults more frequently presented to the ED outside general-practitioner hours, mostly on Sundays. Older persons were more likely to present at any time, rather than specific hours, days or seasons. ED service performance measured against commonly used access-target indicators were rarely satisfied for older people and frequently satisfied for children. Clustering Large Applications is effective in finding latent groups in large-scale mixed-type data, as demonstrated in the present study. Six types of ED presentations were identified and described using clinically relevant characteristics. The present study provides evidence for policy makers in Australia to develop alternative ED models of care tailored around the care needs of the differing groups of patients and thereby supports the sustainable delivery of acute healthcare.

Penggerombolan Sekolah pada Penerimaan Mahasiswa Baru Jalur SNMPTN di IPB Menggunakan Metode Two-Step Cluster

New student admissions are opened in three pathways including SNMPTN, SBMPTN, and Seleksi Mandiri. In order to improve the SNMPTN selection system at IPB, a study was conducted on the quality of SMA/MA which registered to IPB through school clustering. In general, cluster analysis cannot handle large and mixed-type data, so this school clustering used the Two-Step Cluster method with two alternatives, namely without handling outliers and handling 5 percent outliers. Both of these alternatives produced an average Silhouette coefficient value of 0.2 and 0.3 respectively, which was still under the good category. However, clustering without handling outliers resulted in more detailed cluster criteria with 4 optimal clusters. The criteria for these four clusters include, Cluster 1 is a category of Low Commitment, Low Quality, and Low Consistency schools, Cluster 2 and 3 are categories of schools that have special criteria in certain categories, and Cluster 4 is a category of High Commitment, High Quality, and High Consistency.

A clustering procedure for mixed-type data to explore ego network typologies: an application to elderly people living alone in Italy

The analysis of ego networks has attracted a great attention recently and found application in many areas of the social sciences. In particular, the identification of network typologies has become a crucial task and a powerful tool to capture aspects of the social space or personal community in which people are embedded. In this work, we propose a distance-based clustering procedure to identify homogeneous groups of ego networks that are only described by a small number of compositional variables. The proposed approach is motivated by the empirical study of ego networks of contacts extracted from the “Family and Social Subjects” (FSS) Survey conducted by the Italian National Statistical Institute in 2016, which is not specifically oriented to network analysis. We focus on elderly respondents living alone, which can be regarded as a vulnerable category, with the aim to describe their network of contacts. First, mining relational information in FSS data, we derive the ego networks of respondents. Then, we develop a methodology for coping with the presence of heterogeneous data and small amount of information from a network perspective. To this aim, we introduce a dissimilarity measure for mixed-type data, and exploit hierarchical clustering for grouping ego networks according to their composition. In doing so, we intend to make our approach applicable to various surveys.

Latentcor: An R Package for estimating latent correlations from mixed data types

We present `latentcor`, an R package for correlation estimation from data with mixed variable types. Mixed variables types, including continuous, binary, ordinal, zero-inflated, or truncated data are routinely collected in many areas of science. Accurate estimation of correlations among such variables is often the first critical step in statistical analysis workflows. Pearson correlation as the default choice is not well suited for mixed data types as the underlying normality assumption is violated. The concept of semi-parametric latent Gaussian copula models, on the other hand, provides a unifying way to estimate correlations between mixed data types. The R package `latentcor` comprises a comprehensive list of these models, enabling the estimation of correlations between any of continuous/binary/ternary/zero-inflated (truncated) variable types. The underlying implementation takes advantage of a fast multi-linear interpolation scheme with an efficient choice of interpolation grid points, thus giving the package a small memory footprint without compromising estimation accuracy. This makes latent correlation estimation readily available for modern high-throughput data analysis.

Identifying homogeneous subgroups of patients and important features: a topological machine learning approach

BackgroundThis paper exploits recent developments in topological data analysis to present a pipeline for clustering based on Mapper, an algorithm that reduces complex data into a one-dimensional graph.ResultsWe present a pipeline to identify and summarise clusters based on statistically significant topological features from a point cloud using Mapper.ConclusionsKey strengths of this pipeline include the integration of prior knowledge to inform the clustering process and the selection of optimal clusters; the use of the bootstrap to restrict the search to robust topological features; the use of machine learning to inspect clusters; and the ability to incorporate mixed data types. Our pipeline can be downloaded under the GNU GPLv3 license at https://github.com/kcl-bhi/mapper-pipeline.

Subsampling and Aggregation: A Solution to the Scalability Problem in Distance-Based Prediction for Mixed-Type Data

The distance-based linear model (DB-LM) extends the classical linear regression to the framework of mixed-type predictors or when the only available information is a distance matrix between regressors (as it sometimes happens with big data). The main drawback of these DB methods is their computational cost, particularly due to the eigendecomposition of the Gram matrix. In this context, ensemble regression techniques provide a useful alternative to fitting the model to the whole sample. This work analyzes the performance of three subsampling and aggregation techniques in DB regression on two specific large, real datasets. We also analyze, via simulations, the performance of bagging and DB logistic regression in the classification problem with mixed-type features and large sample sizes.