Machine Learning Clustering Methods Research Articles

Regional Spatial Mean of Ionospheric Irregularities Based on K-Means Clustering of ROTI Maps

In this paper, we investigate and propose the application of an unsupervised machine learning clustering method to characterize the spatial and temporal distribution of ionospheric plasma irregularities over the Western African equatorial region. The ordinary Kriging algorithm was used to interpolate the rate of change of the total electron content (TEC) index (ROTI) over gridded 0.5° by 0.5° latitude and longitude regional maps in order to simulate the level of ionospheric plasma irregularities in a quasi-real-time scenario. K-means was used to obtain a spatial mean index through an optimal stratification of regional post-processed ROTI maps. The results obtained could be adapted by appropriate K-means algorithms to a real-time scenario, as has been performed for other applications. This method could allow us to monitor plasma irregularities in real time over the African region and, therefore, lead to the possibility of mitigating their effects on satellite-based location systems in the said region.

Manual Label and Machine Learning in Clustering and Predicting Student Performance

This study presents the Academic Investment Model (AIM) as a novel approach to predicting student academic performance by incorporating learning styles as a predictive feature. Utilizing data from 138 Marketing students across China, the research employs a combination of machine learning clustering methods and manual feature engineering through a four-quadrant clustering technique. The AIM model delineates student investment into four quadrants based on their time and energy commitment to academic pursuits, distinguishing between result-oriented and process-oriented investments. The findings reveal that the four-quadrant method surpasses machine learning clustering in predictive accuracy, highlighting the robustness of manual feature engineering. The study's significance lies in its potential to guide educators in designing targeted interventions and personalized learning strategies, emphasizing the importance of process-oriented assessment in education. Future research is recommended to expand the sample size and explore the integration of deep learning models for validation.

Exploring the application of machine‐learning techniques in the next generation of long‐term hydropower‐thermal scheduling

AbstractThis paper introduces a shape‐based inflow scenarios reduction framework applied in long‐term hydro‐thermal scheduling. This scheduling problem involves strategically managing the limited stored hydro energy in coordination with the electricity system. The scenario fan problem, selected as a long‐term hydro‐thermal scheduling approach, can effectively represent the individual hydropower plants and their short‐term flexibility within the long‐term scheduling process, making it well‐suited for capturing the disaggregated physical characteristics of hydro stations. However, the computational burden of scenario fan problem increases due to the complexity of the cascaded system and the large number of inflow scenarios. To tackle these challenges, a machine learning clustering method is developed for extracting shape‐based features to capture the energy and shape attributes of hydro inflows. A comparative analysis was conducted to evaluate the effectiveness of the proposed method, comparing it with four established clustering methods. This evaluation was carried out on a hydro‐thermal test case, utilizing both in‐sample and out‐of‐sample data analysis. The findings demonstrate that the proposed framework outperforms other methods in terms of computational time by 42.6% and optimal reservoir trajectories by 12.1% considering K‐means using Euclidean distance as the benchmark. Additionally, this study showcases the feasibility of performing long‐term hydro‐thermal scheduling at a granular level, with detailed topological information and a shorter time scale. This expanded approach opens up new possibilities for addressing the integration of renewable energy resources within the realm of long‐term hydro‐thermal scheduling.

How do medical professionals make sense (or not) of AI? A social-media-based computational grounded theory study and an online survey

To investigate opinions and attitudes of medical professionals towards adopting AI-enabled healthcare technologies in their daily business, we used a mixed-methods approach. Study 1 employed a qualitative computational grounded theory approach analyzing 181 Reddit threads in the several subreddits of r/medicine. By utilizing an unsupervised machine learning clustering method, we identified three key themes: (1) consequences of AI, (2) physician–AI relationship, and (3) a proposed way forward. In particular Reddit posts related to the first two themes indicated that the medical professionals' fear of being replaced by AI and skepticism toward AI played a major role in the argumentations. Moreover, the results suggest that this fear is driven by little or moderate knowledge about AI. Posts related to the third theme focused on factual discussions about how AI and medicine have to be designed to become broadly adopted in health care. Study 2 quantitatively examined the relationship between the fear of AI, knowledge about AI, and medical professionals' intention to use AI-enabled technologies in more detail. Results based on a sample of 223 medical professionals who participated in the online survey revealed that the intention to use AI technologies increases with increasing knowledge about AI and that this effect is moderated by the fear of being replaced by AI.

A machine learning framework to assess the impact of the COVID-19 pandemic on electricity consumption patterns

Energy self-sufficiency and resilience are essential elements for long-term urban sustainability. Energy supply in many cities in the United States has been highly dependent on fossil fuels and excessive energy consumption, both at home and outside (i.e., residential and non-residential energy consumption). In this work, we identify electricity consumption spatial trends and profiles across neighborhoods in the city of Chicago using a self-organizing map (SOM) machine learning method as a main tool. Toward this goal, we use an anonymous dataset provided by Commonwealth Edison (ComEd; Chicago's main electricity provider), which includes electricity consumption data of individual residential and non-residential accounts in Chicago. We aggregate electricity consumption to the zip codes level, apply machine learning clustering methods to categorize electricity consumption patterns, and finally couple it with land use variables to find spatial interrelations. The output demonstrates how specific profiles of electricity consumption emerge across zip codes, which is the first step for developing tailored policies to lower energy consumption. The result of this study can help engineers, urban planners, and policymakers develop actionable strategies for electricity consumption management.

A Fuzzy Clustering Approach to Identify Pedestrians' Traffic Behavior Patterns.

Pattern recognition of pedestrians' traffic behavior can enhance the management efficiency of interested groups by targeting access to them and facilitating planning via more specific surveys. This study aimed to evaluate the pedestrians' traffic behavior pattern by fuzzy clustering algorithm and assess the factors related to higher-risk traffic behavior of pedestrians. Study Design: This study is a secondary methodological study based on the data from a cross-sectional study. The fuzzy c-means (FCM), as a machine learning clustering method, was conducted to identify the pattern of traffic behaviors by collecting data from 600 pedestrians in Urmia, Iran via "the Pedestrian Behavior Questionnaire" (PBQ) and using 5 domains of PBQ. Multiple logistic regression was fitted to identify risk factors of traffic behaviors. Results revealed two clusters consisting of lower-risk and higher-risk behaviors. The majority of pedestrians (64.33%) were in the lower-risk cluster. Subjects≤33 years old (Odds ratio [OR]=1.92, P<0.001), subjects with≤6 years of education (OR=1.74, P=0.010), males (OR=1.90, P=0.001), unmarried pedestrians (OR=3.61, P=0.007), and users of public transportation (OR=2.01, P=0.002) were more likely to have higher-risk traffic behavior. We identified traffic behavior patterns of Urmia pedestrians with lower-risk and higher-risk behaviors via FCM. The findings from this study would be helpful for policymakers to promote safety measures and train pedestrians.

Reconfiguration of last-mile supply chain for parcel delivery using machine learning and routing optimization

Last-mile delivery has several negative environmental impacts in urban areas because of its high levels of greenhouse gas emissions and air pollution, as well as traffic congestion. These issues motivate decision-makers to redesign the delivery networks and make them more sustainable and efficient. A well-planned territory design can reduce total travel times and distances in urban distribution systems, in addition to balancing the workload between drivers. In this study, a two-echelon parcel distribution network modeled as the two-echelon vehicle routing problem with territory design and satellite location decisions is considered. A three-stage decomposition algorithm is proposed to solve this problem. In the first stage, a non-supervised machine learning clustering method is applied, followed by an algorithm based on the nearest-neighbor routing procedure, to find a set of routes for the second and first echelons. An improvement heuristic was also applied to improve the results in terms of the second echelon routing, considering the computational complexity of a large-scale instance. A case study based on real data from a delivery company in the city of Paris, France is adopted to perform the experiments. The outcomes of this paper show an improvement of 22.6% in travel time and distance. This reduction is also assessed with performance indicators like land use, fixed costs, energy consumption, carbon dioxide equivalent, and fine particles emissions.

Hyperparameter optimization for cardiovascular disease data-driven prognostic system

Prediction and diagnosis of cardiovascular diseases (CVDs) based, among other things, on medical examinations and patient symptoms are the biggest challenges in medicine. About 17.9 million people die from CVDs annually, accounting for 31% of all deaths worldwide. With a timely prognosis and thorough consideration of the patient’s medical history and lifestyle, it is possible to predict CVDs and take preventive measures to eliminate or control this life-threatening disease. In this study, we used various patient datasets from a major hospital in the United States as prognostic factors for CVD. The data was obtained by monitoring a total of 918 patients whose criteria for adults were 28-77 years old. In this study, we present a data mining modeling approach to analyze the performance, classification accuracy and number of clusters on Cardiovascular Disease Prognostic datasets in unsupervised machine learning (ML) using the Orange data mining software. Various techniques are then used to classify the model parameters, such as k-nearest neighbors, support vector machine, random forest, artificial neural network (ANN), naïve bayes, logistic regression, stochastic gradient descent (SGD), and AdaBoost. To determine the number of clusters, various unsupervised ML clustering methods were used, such as k-means, hierarchical, and density-based spatial clustering of applications with noise clustering. The results showed that the best model performance analysis and classification accuracy were SGD and ANN, both of which had a high score of 0.900 on Cardiovascular Disease Prognostic datasets. Based on the results of most clustering methods, such as k-means and hierarchical clustering, Cardiovascular Disease Prognostic datasets can be divided into two clusters. The prognostic accuracy of CVD depends on the accuracy of the proposed model in determining the diagnostic model. The more accurate the model, the better it can predict which patients are at risk for CVD.

Urban Logistics through River: A Two-Echelon Distribution Model

Studies that use rivers in a last-mile delivery context are scarce. This research considers the first multimodal alternative based on a barge for parcel delivery activities. It proposes two sustainable network designs for a two-echelon distribution. The efficiency of scenarios is assessed through performance indicators. A three-stage decomposition heuristic is used. Allocation of the customers to the closest satellite at the first stage uses a non-supervised machine learning clustering method, 2D-k-means. The last two stages, comprising the two echelons routing, are solved using a heuristic based on the nearest neighbor procedure. The fixed costs decrease by 41% and energy consumption by 92% when applying a river transportation mode and e-cargo bikes in the distribution network’s first and second echelon, respectively. Future research avenues are to render the results more realistic with the consideration of other costs and a larger network.

Exploring the potential role of bikeshare to complement public transit: The case of San Francisco amid the coronavirus crisis

The recent worldwide SARS-CoV-2 (COVID-19) pandemic has reshaped the way people live, how they access goods and services, and how they perform various activities. For public transit, there have been health concerns over the potential spread to transit users and transit service staff, which prompted transportation agencies to make decisions about the service, e.g., whether to reduce or temporarily shut down services. These decisions had substantial negative consequences, especially for transit-dependent travelers, and prompted transit users to explore alternative transportation modes, e.g., bikeshare. However, local governments and the public in general have limited information about whether and to what extent bikeshare provides adequate accessibility and mobility to those transit-dependent residents. To fill this gap, this study implemented spatial and visual analytics to identify how micro-mobility in the form of bikesharing has addressed travel needs and improved the resilience of transportation systems. The study analyzed the case of San Francisco in California, USA, focusing on three phases of the pandemic, i.e., initial confirmed cases, shelter-in-place, and initial changes in transit service. First, the authors implemented unsupervised machine learning clustering methods to identify different bikesharing trip types. Moreover, through spatiotemporally matching bikeshare ridership data with transit service information (i.e., General Transit Feed Specification, GTFS) using the tool called OpenTripPlanner (OTP), the authors studied the travel behavior changes (e.g., the proportion of bikeshare trips that could be finished by transit) for different bikeshare trip types over the three specified phases. This study revealed that during the pandemic, more casual users joined bikeshare programs; the proportion of recreation-related bikeshare trips increased; and routine trips became more prevalent considering that docking-station-based bikeshare trips increased. More importantly, the analyses also provided insights about mode substitution, because the analyses identified an increase in dockless bikeshare trips in areas with no or limited transit coverage.

Carpooling: Who is closest to adopting it? An investigation into the potential car-poolers among private vehicle users: A case of a developing country, India

The demand for transit is linked to a country's economic and population growth. As per the Ministry of Road Transport and Highways (MORTH), from 2009 to 2019, India witnessed an average growth rate of 9.9% per annum in registered vehicles. Even though the vehicle ownership is often viewed as a symbol of economic prosperity, the associated negative externalities (congestion, air pollution, road crashes, noise pollution, etc.) are substantial. To counter this, carpooling has been proposed as a solution globally. However, there is a paucity of literature for analysing the difference in travel characteristics of car-poolers among other private vehicle users. The present study aims to bridge this gap by comparing the demographic and travel characteristics of existing car-poolers with two-wheelers and single-occupant car users in Gurugram in the national capital region of India. Using an unsupervised machine learning clustering method (two-step cluster), the findings of the present study indicate that the profile of single-occupant car (SOC) users and car-poolers are highly distinguished from two-wheeler users. However, car-poolers have similar characteristics to single-occupant car users, except for travel cost, income level, and gender, suggesting that SOC users are potential car-poolers. A comparative study of this kind would help city planners and policymakers identify prospective car-poolers, and formulate policies to encourage SOC users to adopt carpooling and promote urban transportation sustainability.

Revisiting Atmospheric Features of Mars Orbiter Laser Altimeter Data Using Machine Learning Algorithms

AbstractThe Mars Orbiter Laser Altimeter (MOLA) instrument has been drawing a map of Mars' topography between September 1997 and June 2001. It has also been able to observe clouds during the mission duration, providing data for the low Martian atmosphere for nearly 1.5 Mars years. The Mars Global Surveyor, which carried MOLA, also carried two other instruments that also observed clouds during the same time period (the Mars Orbiter Camera and the Thermal Emission Spectrometer). Combining observations from these three data sets could provide a complete recap of most atmospheric structures during MY24 and MY25. However, previous studies of MOLA data set often had to use stringent detection criteria. Using machine learning clustering methods, we end up finding way more atmospheric returns. Our results are presented in the form of an atmospheric features catalog that we then use to compare MOLA observations with Mars Orbiter Camera and Thermal Emission Spectrometer results, but also with more recent missions. We study the development of recurrent phenomenon in the Martian atmosphere, like the aphelion cloud belt or the south polar hood, but also spontaneous events such as regional dust storms. Methods could be tuned even more finely by using more complex clustering methods or deep learning algorithms to clearly distinguish atmospheric structures.

Survival analysis of patient groups defined by unsupervised machine learning clustering methods based on patient metabolomic data.

PurposeMeta-analyses failed to accurately identify patients with non-metastatic breast cancer who are likely to benefit from chemotherapy, and metabolomics could provide new answers. In our previous published work, patients were clustered using five different unsupervised machine learning (ML) methods resulting in the identification of three clusters with distinct clinical and simulated survival data. The objective of this study was to evaluate the survival outcomes, with extended follow-up, using the same 5 different methods of unsupervised machine learning. Experimental designForty-nine patients, diagnosed between 2013 and 2016, with non-metastatic BC were included retrospectively. Median follow-up was extended to 85.8 months. 449 metabolites were extracted from tumor resection samples by combined Liquid chromatography-mass spectrometry (LC–MS). Survival analyses were reported grouping together Cluster 1 and 2 versus cluster 3. Bootstrap optimization was applied. ResultsPCA k-means, K-sparse and Spectral clustering were the most effective methods to predict 2-year progression-free survival with bootstrap optimization (PFSb); as bootstrap example, with PCA k-means method, PFSb were 94% for cluster 1&2 versus 82% for cluster 3 (p = 0.01). PCA k-means method performed best, with higher reproducibility (mean HR=2 (95%CI [1.4–2.7]); probability of p ≤ 0.05 85%). Cancer-specific survival (CSS) and overall survival (OS) analyses highlighted a discrepancy between the 5 ML unsupervised methods. ConclusionOur study is a proof-of-principle that it is possible to use unsupervised ML methods on metabolomic data to predict PFS survival outcomes, with the best performance for PCA k-means. A larger population study is needed to draw conclusions from CSS and OS analyses.

An unsupervised machine learning approach using passive movement data to understand depression and schizophrenia

IntroductionSchizophrenia and Major Depressive Disorder (MDD) are highly burdensome mental disorders, with significant cost to both individuals and society. Despite these disorders representing distinct clinical categories, they are each heterogenous in their symptom profiles, with considerable transdiagnostic features. Although movement and sleep abnormalities exist in both disorders, little is known of the precise nature of these changes longitudinally. Passively-collected longitudinal data from wearable sensors is well suited to characterize naturalistic features which may cross traditional diagnostic categories (e.g., highlighting behavioral markers not captured by self-report information). MethodsThe present analyses utilized raw minute-level actigraphy data from three diagnostic groups: individuals with schizophrenia (N = 23), individuals with depression (N = 22), and controls (N = 32), respectively, to interrogate naturalistic behavioral differences between groups. Subjects' week-long actigraphy data was processed without diagnostic labels via unsupervised machine learning clustering methods, in order to investigate the natural bounds of psychopathology. Further, actigraphic data was analyzed across time to determine timepoints influential in model outcomes. ResultsWe find distinct actigraphic phenotypes, which differ between diagnostic groups, suggesting that unsupervised clustering of naturalistic data aligns with existing diagnostic constructs. Further, we found statistically significant inter-group differences, with depressed persons showing the highest behavioral variability. LimitationsHowever, diagnostic group differences only consider biobehavioral trends captured by raw actigraphy information. ConclusionsPassively-collected movement information combined with unsupervised deep learning algorithms shows promise in identifying naturalistic phenotypes in individuals with mental health disorders, specifically in discriminating between MDD and schizophrenia.

Clustering of facies in tight carbonates using machine learning

Machine learning clustering methods offer the potential for recognition and separation of facies based on core or well-log data. This is a particular problem for carbonate rocks because diagenesis produces a wide range of rock microstructures and transport properties. In this work we use a large database of high quality poroperm, electrical, mercury injection capillary pressure and nuclear magnetic resonance spectroscopy measurements (307 core samples), representing 5 stratigraphically defined facies, as well as well log data to examine facies-recognition abilities using 8 different machine learning clustering approaches and a redundancy of 10 to ensure statistically valid results, resulting in a total of over 990 clustering runs. For a 3 cluster problem, we find that the Expectation Maximisation (92.57% success) and two types of Kmeans approaches (89.60% and 91.09%) provide the best methods. Further testing using the best of these shows that the quality of the input parameter (attribute) matters more than the number of attributes used, with the power of attributes in decreasing order being porosity, cementation exponent, permeability, pore throat diameter and free fluid index, implying that some attributes can degrade clustering performance. Further tests show that there should be at least as many attributes as clusters, in which case the machine learning can be left to choose the final number of clusters, providing the best performance in this work (69.35% success for a five cluster problem), otherwise it is best to constrain the cluster number by supervision. Application of the results from the previous testing to a mixed carbonate tight carbonate well from the Butmah formation shows satisfactory determination of 4 petrofacies by clustering (up to 91.65%) when compared to petrofacies determined manually. However, the greater challenge of clustering 9 reservoir quality classes defined using a ternary petrofacies approach did not provide a successful result (<38% success rate).

Process-Oriented Estimation of Chlorophyll-a Vertical Profile in the Mediterranean Sea Using MODIS and Oceanographic Float Products

Reconstructing chlorophyll-a (Chl-a) vertical profile is a promising approach for investigating the internal structure of marine ecosystem. Given that the process of profile classification in current process-oriented profile inversion methods are either too subjective or too complex, a novel Chl-a profile reconstruction method was proposed incorporating both a novel binary tree profile classification model and a profile inversion model in the Mediterranean Sea. The binary tree profile classification model was established based on a priori knowledge provided by clustering Chl-a profiles measured by BGC-Argo floats performed by the profile classification model (PCM), an advanced unsupervised machine learning clustering method. The profile inversion model contains the relationships between the shape-dependent parameters of the nonuniform Chl-a profile and the corresponding Chl-a surface concentration derived from satellite observations. According to quantitative evaluation, the proposed profile classification model reached an overall accuracy of 89%, and the mean absolute percent deviation (MAPD) of the proposed profile inversion model ranged from 12%–37% under different shape-dependent parameters. By generating monthly three dimensions Chl-a concentration from 2011 to 2018, the proposed process-oriented method exhibits great application potential in investigating the spatial and temporal characteristics of Chl-a profiles and even the water column total biomass throughout the Mediterranean Sea.

Application of Modern Urban Landscape Design Based on Machine Learning Model to Generate Plant Landscaping

With the continuous improvement of the living standards of people, the requirements of people for the environment in which they are located are gradually increasing, which makes urban landscape design work more and more important. Plant landscaping is the core component of urban landscape design. By analyzing and discussing the application of plant landscaping in urban landscape design, the quality level of urban landscape design can be further improved, thereby promoting the sustainable development of urban construction. Machine learning makes it possible to realize the intelligent processing of data and make full use of the knowledge and value contained in the data. This paper explores the way of intelligent analysis and application based on machine learning in the field of landscape architecture. First, the k-means machine learning clustering method is used to determine the types of plant landscaping required by modern urban landscape design. Experiments show that our model can be well applied to design applications of modern urban landscapes.

Applying adaptive learning by integrating semantic and machine learning in proposing student assessment model

&lt;p&gt;Adaptive learning is one of the most widely used data driven approach to teaching and it received an increasing attention over the last decade. It aims to meet the student’s characteristics by tailoring learning courses materials and assessment methods. In order to determine the student’s characteristics, we need to detect their learning styles according to visual, auditory or kinaesthetic (VAK) learning style. In this research, an integrated model that utilizes both semantic and machine learning clustering methods is developed in order to cluster students to detect their learning styles and recommend suitable assessment method(s) accordingly. In order to measure the effectiveness of the proposed model, a set of experiments were conducted on real dataset (Open University Learning Analytics Dataset). Experiments showed that the proposed model is able to cluster students according to their different learning activities with an accuracy that exceeds 95% and predict their relative assessment method(s) with an average accuracy equals to 93%.&lt;/p&gt;

Determination of miscible CO2 flooding analogue projects with machine learning

Using analogue reservoirs for comparison and benchmarking is a comprehensive method for ensuring accurate measurement and gaining a better understanding of potential oil recovery enhancement prospects. The approach of leveraging information from analogue reservoirs is extended to CO 2 EOR project management in this study. We present a novel application of machine learning clustering algorithms to the rapid identification of analogues for new projects without having to sift through massive amounts of data. We use machine learning clustering methods to group successfully executed miscible CO 2 flooding projects into clusters of projects with similar fluid/reservoir characteristics and to identify analogues for new target projects. Porosity, permeability, oil gravity and viscosity, reservoir pressure and temperature, minimum miscibility pressure (MMP), and depth were all input parameters. Data from nearly 200 miscible CO 2 EOR projects around the world were clustered using the Agglomerative Hierarchical Clustering Algorithm (HCA), K-Means, and K-Median techniques. To reduce information redundancy in high-dimensional data, Principal Component Analysis was used as a pre-processing step. Three evaluation indices (the Davies Bouldin, Calinski Harabasz, and Silhouette Coefficient Scores) were used to compare the efficiency of the clustering algorithms and choose the best one for this dataset. A Principal Component – weighted Euclidean distance similarity metric was computed using three existing miscible CO 2 flooding projects (Weyburn, Hansford Marmaton, and Paradis) as test cases to confirm the clustering results. The clustering analysis identified five different classes of miscible CO 2 projects, each with its reservoir and fluid characteristics. Type 1 projects, in general, are those that are carried out primarily in shallow carbonate reservoirs at the lowest temperatures and pressures of any database project with typical porosity and permeability. More than 60% of Type 2 projects are in sandstone reservoirs. They are at shallower depths with lower temperatures, pressures, porosities, and permeabilities than project Type 4. Type 3 projects are typically undertaken in carbonate reservoirs with medium depths and temperatures but the highest reservoir pressures. This project type has medium porosity and permeability. In comparison to project Type 2, Type 4 primarily consists of projects conducted in sandstone formations at great depths with high reservoir temperatures and pressures. The porosity and permeability of these project types are average. Finally, Type 5 projects are typically undertaken in sandstone formations at average depths, temperatures, and pressures, but with the greatest porosity and permeability of all project types. In addition to the clustering analysis, the distance similarity metric used in this work identified projects that were most like the test miscible CO 2 flooding project cases. Key rock and fluid properties, well types, and best infill drilling strategies, recovery improvement strategies, and production performance can all be learned from identified analogue projects. This data can be used to improve the operational, technical, field, and well-planning decisions for new CO 2 flooding projects. The workflow demonstrated in this paper is easily adaptable to data sets from other flooding projects. • K-Means clustering model can be used effectively to categorize miscible CO 2 flooding projects. • The global miscible CO 2 flooding projects in collated dataset can be broadly divided into five classes by K-means clustering model. • A calculated Principal Component – weighted Euclidean distance similarity metric can be used to successfully validate cluster groupings. • Ideal analogue projects with similar fluid and reservoir properties can be identified with K-Means and the distance-based metric

Validation of cardiogenic shock phenotypes in a mixed cardiac intensive care unit population.

Proposed phenotypes have recently been identified in cardiogenic shock (CS) populations using unsupervised machine learning clustering methods. We sought to validate these phenotypes in a mixed cardiac intensive care unit (CICU) population of patients with CS. We included Mayo Clinic CICU patients admitted from 2007 to 2018 with CS. Agnostic K means clustering was used to assign patients to threeclusters based on admission values of estimated glomerular filtration rate, bicarbonate, alanine aminotransferase, lactate, platelets, and white blood cell count. In-hospital mortality and 1-year mortality were analyzed using logistic regression and Cox proportional-hazards models, respectively. We included 1498 CS patients with a mean age of 67.8 ± 13.9 years, and 37.1% were females. The acute coronary syndrome was present in 57.3%, and cardiac arrest was present in 34.0%. Patients were assigned to clusters as follows: Cluster 1 (noncongested), 603 (40.2%); Cluster 2 (cardiorenal), 452 (30.2%); and Cluster 3 (hemometabolic), 443 (29.6%). Clinical, laboratory, and echocardiographic characteristics differed across clusters, with the greatest illness severity in Cluster 3. Cluster assignment was associated with in-hospital mortality across subgroups. In-hospital mortality was higher in Cluster 3 (adjusted odds ratio [OR]: 2.6 vs. Cluster 1 and adjusted OR: 2.0 vs. Cluster 2, both p < 0.001). Adjusted 1-year mortality was incrementally higher in Cluster 3 versus Cluster 2 versus Cluster 1 (all p < 0.01). We observed similar phenotypes in CICU patients with CS as previously reported, identifying a gradient in both in-hospital and 1-year mortality by cluster. Identifying these clinical phenotypes can improve mortality risk stratification for CS patients beyond standard measures.