Unsupervised Machine Learning Model Research Articles

Assessing and predicting green gentrification susceptibility using an integrated machine learning approach

ABSTRACT Greenery initiatives, such as green infrastructures (GIs), create sustainable and climate-resilient environments. However, they can also have unintended consequences, such as displacement and gentrification in low-income areas. This paper proposes an integrated machine learning (ML) approach that combines both unsupervised and supervised ML algorithms. First, 35 indicators that contribute to green gentrification were identified and categorised into 7 categories: social, economic, demographic, housing, household, amenities, and GIs. Second, data was collected for all census tracts in New York City. Third, the green gentrification susceptibility was modelled into 6 levels using k-means clustering analysis, which is an unsupervised ML model. Fourth, the Technique for Order of Preference by Similarity to the Ideal Solution (TOPSIS) was used to map the census tracts to their green gentrification susceptibility level. Finally, different supervised ML algorithms were trained and tested to predict the green gentrification susceptibility. The results showed that the artificial neural network (ANN) model is the most accurate in classifying and predicting the green gentrification susceptibility with an overall accuracy of 96%. Moreover, the outcomes showed that the Normal Difference Vegetation Index (NDVI), the proximity to GIs, the GIs frequency, and the total area of GIs were identified as the most important indicators to predict green gentrification susceptibility. Ultimately, the proposed approach allows practitioners and researchers to perform micro-level (i.e. on the census-tracts level) predictions and inferences about green gentrification susceptibility. This allows more focused and targeted mitigation actions to be designed and implemented in the most affected communities, thus promoting environmental justice.

Extracting Features from Satellite Imagery to Understand the Size and Scale of Housing Sub-Markets in Madrid

The following paper proposes a novel machine learning approach to the segmentation of urban housing markets. We extract features from globally available satellite imagery using an unsupervised machine learning model called MOSAIKS, and apply a k-means clustering algorithm to the extracted features to identify sub-markets at multiple intra-urban scales within a case study of Madrid (Spain). To systematically explore scale effects on the resulting clusters, the analysis is repeated with varying sizes of satellite image patches. We assess the resulting clusters across scales using several internal cluster-evaluation metrics. Additionally, we use data from online listings portal Idealista to measure the homogeneity of housing prices within the clusters, to understand how well sub-markets can be differentiated by the image features. This paper evaluates the strengths and weakness of the method to identify urban housing sub-markets, a task which is important for planners and policy makers and is often limited by a lack of data. We conclude that the approach seems useful to divide large urban housing markets according to different attributes and scales.

Feeling analysis on allergen immunotherapy on Twitter using an unsupervised machine learning model

Analyze feelings about allergen-specific immunotherapy on Twitter using the VADER model VADER (Valence Aware Dictionary and sEntiment Reasoner) model. tweets related to specific allergen immunotherapy were obtained through the Twitter Application Programming Interface (API). The keywords "allergy shot" were used between January 1, 2012, and December 31, 2022. The data was processed by removing URLs, usernames, hashtags, multiple spaces, and duplicate tweets. Subsequently, a sentiment analysis was performed using the VADER model. A total of 34,711 tweets were retrieved, of which 1928 were eliminated. Of the remaining 32,783 tweets, 32.41% expressed a negative sentiment, 31.11% expressed a neutral sentiment, and 36.47% expressed a positive sentiment, with an average polarity of 0.02751 (neutral) over the 11-year period. The average polarity of tweets about allergen-specific immunotherapy is neutral over the 11 years analyzed. There was an annual increase in the average polarity over the years, with 2017, 2018, and 2022 having positive polarity averages. Additionally, the number of tweets decreased over time.

Classification of laser beam profiles using machine learning at the ELI-NP high power laser system

The high power laser system at Extreme Light Infrastructure—Nuclear Physics has demonstrated 10 PW power shot capability. It can also deliver beams with powers of 1 PW and 100 TW in several different experimental areas that carry out dedicated sets of experiments. An array of diagnostics is deployed to characterize the laser beam spatial profiles and to monitor their evolution during the amplification stages. Some of the essential near-field and far-field profiles acquired with CCD cameras are monitored constantly on a large screen television for visual observation and for decision making concerning the control and tuning of the laser beams. Here, we present results on the beam profile classification obtained from datasets with over 14 600 near-field and far-field images acquired during two days of laser operation at 1 PW and 100 TW. We utilize supervised and unsupervised machine learning models based on trained neural networks and an autoencoder. These results constitute an early demonstration of machine learning being used as a tool in the laser system data classification.

Abstract 4912: Unsupervised detection of stromal phenotypes with distinct fibrogenic and inflamed properties in NSCLC

Abstract Background: Understanding the composition of cancer-associated stroma (CAS) is vital, as the number and location of immune cells and fibroblasts, as well as the degree of extracellular matrix deposition, have implications for cancer progression and response to treatment, including in non-small cell lung cancer (NSCLC). Manual analysis of CAS does not fully describe the stromal milieu, especially from a spatial perspective, and is highly subjective. To this end, we have developed an unsupervised machine learning (ML) model to characterize the CAS in NSCLC from hematoxylin and eosin (H&E) stained whole slide images (WSI) at scale. Methods: PathExploreTM models were deployed to predict stromal tissue and cell types, while another ML model was used to detect collagen fibers from H&E stained WSIs from the TCGA LUAD (N=536) and LUSC (N=464) datasets. Stroma was divided into small regions (median = 0.02 mm2), and 88 features characterizing cell distribution, tissue composition and fiber density were extracted from each region. Graphs were generated connecting neighboring regions (nodes), and an unsupervised variational graph auto-encoder (VGAE) model was trained to learn 8 latent features through dimensionality reduction. Stromal phenotypes were then derived from the latent features using k-means clustering. The fraction of each phenotype in the stroma was correlated against immune- and stroma-related gene expression signatures (GES) and overall survival (OS). Results: Deployment of VGAE on LUAD and LUSC WSIs revealed three distinct stromal phenotypes - P0, P1 and P2. Fibroblast density was elevated in P0 and P1 regions (p<0.001), immune cell density was elevated in P2 regions (p<0.001), and collagen fiber intensity was highest in P1 regions (p<0.001). P2 enrichment was correlated with elevated expression of the T cell-inflamed gene expression profile (TGEP; Spearman ρ = 0.43 in LUAD; ρ = 0.27 in LUSC) and with improved OS (HR = 0.696; 95% CIs: 0.571-0.847 in LUSC). Conversely, P1 enrichment was positively associated with a transforming growth factor-β-induced cancer associated fibroblast GES (TGFβ-CAF: ρ = 0.19 in LUAD and ρ = 0.12 in LUSC) and poor OS (HR = 1.358; 95% CIs: 1.149-1.603 in LUSC). These phenotypes are consistent with fibroblast-enriched, collagen-depleted stroma (P0), collagen-rich, fibroblast-enriched tumor-promoting stroma (P1), and immune cell-enriched, tumor-suppressive stroma (P2). Conclusions: We describe an unsupervised, data-driven method of predicting stromal regions with discrete patterns of cell composition and collagen deposition in NSCLC. This approach identified three phenotypes of NSCLC stroma. These results highlight the ability of ML models to characterize and find meaningful patterns within the cell, tissue, and matrix components of a tumor. This work provides further evidence of the potential of ML to discover novel precision medicine biomarkers in NSCLC. Citation Format: Neel Patel, Nhat Le, Tan Nguyen, Fedaa Najdawi, Sandhya Srinivasan, Adam Stanford-Moore, Deeksha Kartik, Jun Zhang, Jacqueline Brosnan-Cashman, Robert Egger, Justin Lee, Matthew Bronnimann. Unsupervised detection of stromal phenotypes with distinct fibrogenic and inflamed properties in NSCLC [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4912.

Abstract 868: A novel unsupervised machine learning model applied to neuroblastoma single-cell RNA-seq data reveals a drug-induced mesenchymal-like gene expression program

Abstract High-risk neuroblastoma is one of the most difficult to treat pediatric cancers, with a survival rate of only about 50% and significant long-term consequences from current chemotherapy. Preclinical models such as cell lines and mice are the backbone of drug development and experimental-mechanistic oncology, but the development of new treatments is hampered in part by a lack of understanding of the direct clinical relevance of such data collected in preclinical models. Despite this, few formal methods have been developed to determine how these various models represent/resemble primary patient tumors. Here, we present the first comprehensive single-cell RNA-seq analysis of neuroblastoma across an extensive cohort of patient tumors and a variety of preclinical model systems (n = 126 total samples assembled - the largest cohort of its kind). By building an innovative unsupervised machine learning method, which we term “automatic consensus nonnegative matrix factorization” (acNMF), we have integrated and contrasted the transcriptional landscapes of patient tumors with those of cell lines, patient-derived xenografts (PDX), and genetic mouse models (GEMM).Using these tools, we discovered the dominant adrenergic gene expression programs found in neuroblastoma patient tumors were preserved across all preclinical models. However, the presumptive chemo-resistant mesenchymal-like programs, while identifiable in cell lines, were primarily restricted to subpopulations of cancer-associated fibroblasts and Schwann-like cells in vivo. Surprisingly, a mesenchymal-like program could be acutely chemotherapy-induced in GEMM and was evident in pre-treated patient and PDX samples, suggesting a previously uncharacterized mechanism of therapy escape. In addition to these core findings, our computational tools were able to further delineate the classical neuroblastoma adrenergic and mesenchymal gene expression programs, discovering for example, novel subpopulations of cancer associated fibroblasts. These behaviors were conserved across tumors and most preclinical models, which we validated by RNA in situ hybridization, which is a high resolution ultra sensitive spatial transcriptomics technology. Our work cautions against overreliance on traditional preclinical models without recognizing their limitations and we offer a nuanced, high-resolution view of neuroblastoma pre-clinical systems for advancing therapeutic development. We have launched an open-source web resource, featuring this integrated map to aid the scientific community in further exploration of these data and hypothesis generation (available at http://pscb.stjude.org). Citation Format: Rich Chapple, Xueying Liu, Sivaraman Natarajan, Margaret I. Alexander, Yuna Kim, Anand Patel, Christy W. LaFlamme, Min Pan, William C. Wright, Hyeong-Min Lee, Yinwen Zhang, Meifen Lu, Selene C. Koo, Courtney Long, John Harper, Chandra Savage, Melissa D. Johnson, Thomas Confer, Walter J. Akers, Michael A. Dyer, Heather Sheppard, John Easton, Paul Geeleher. A novel unsupervised machine learning model applied to neuroblastoma single-cell RNA-seq data reveals a drug-induced mesenchymal-like gene expression program [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 868.

Improve Unsupervised Machine Learning Model on Fruits by Using VGG-16

The demand for efficient fruit classification has risen with supply chain complexities. For the sake of saving the labor force and improving efficiency, applying machine learning algorithms on fruit categorization to multiple stages in the factory is a feasible solution. The paper briefly delivers the rationale of algorithm VGG-16 and validates its advantage in accuracy by comparing it with other convolutional neural network models. Traditional convolutional network reaches its peak accuracy which is around 40% after 45 training epochs. VGG-16 results with a 97% percent accuracy on the classification of more than 20 kinds of fruits only after 6 epochs and can be improved further by enriching and augmenting the dataset. However, although VGG-16 only requires a small number of epochs, the weighted parameter for each layer increases dramatically and increases the running time. Future researchers should focus on optimizing the algorithm to make it more feasible in the industry. Also, remote computing might be a solution for large computational requirements.

Psychophysiological strategies for enhancing performance through imagery–skin conductance level analysis in guided vs. self-produced imagery

Athletes need to achieve their optimal level of arousal for peak performance. Visualization or mental rehearsal (i.e., Imagery) often helps to obtain an appropriate level of activation, which can be detected by monitoring Skin Conductance Level (SCL). However, different types of imagery could elicit different amount of physiological arousal. Therefore, this study aims: (1) to investigate differences in SCL associated with two instructional modalities of imagery (guided vs. self-produced) and six different scripts; (2) to check if SCL could differentiate respondents according to their sport expertise. Thirty participants, aged between 14 and 42 years (M = 22.93; SD = 5.24), with different sport levels took part in the study. Participants listened to each previously recorded script and then were asked to imagine the scene for a minute. During the task, SCL was monitored. We analysed the mean value, variance, slope and number of fluctuations per minute of the electrodermal signal. Unsupervised machine learning models were used for measuring the resemblance of the signal. The Wilcoxon signed-rank test was used for distinguishing guided and self-produced imagery, and The Mann–Whitney U test was used for distinguishing results of different level athletes. We discovered that among others, self-produced imagery generates lower SCL, higher variance, and a higher number of fluctuations compared to guided imagery. Moreover, we found similarities of the SCL signal among the groups of athletes (i.e. expertise level). From a practical point of view, our findings suggest that different imagery instructional modalities can be implemented for specific purposes of mental preparation.

The role of Passion and Self-Efficacy in entrepreneurial activities in the gig economy: An Unsupervised Machine Learning Analysis with Topic Modeling

This research examines passion and self-efficacy through experience and knowledge, as motivational factors that support entrepreneurs within the gig economy (GE). It sheds light on entrepreneurs’ sources of passion in the GE literature. The sample is composed of all the 1164 entrepreneurship activities offered worldwide through Airbnb, Tour by Locals, and Withlocals on 5 May 2022. The study is supported by unsupervised machine learning models and seeks to find latent topics emerging from the analysis of the entrepreneurs’ descriptions and exposes the main correlated clustered dimensions. There are six main motivators behind GE platforms as a first step toward entrepreneurship: Experience, Passion for share, Knowledge, Classic traditions, Empowered community and local activities, and Well-being. It also confirms the correlation between passion and self-efficacy through experience, pointing them as main factors behind entrepreneurship in the GE. Five of the six sources of passion previously pointed by theory were found: Passion for growth, for people, for product/service, for innovation and for social mission. This study discloses self-efficacy and the sources of passion and points directions to practitioners involved in entrepreneurial activities in the GE ecosystem. This work used machine learning models to access quantitatively a paradigm that is inductive by nature. The results point to well-being as a significant factor to be addressed in future research regarding entrepreneurship. This research only studies individuals involved in the GE; as such, further studies should cohort new populations from different fields.

Avoiding food waste from restaurant tickets: a big data management tool

PurposeThis study aims to address the global food waste problem in restaurants by analyzing customer sales information provided by restaurant tickets to gain valuable insights into directing sales of perishable products and optimizing product purchases according to customer demand.Design/methodology/approachA system based on unsupervised machine learning (ML) data models was created to provide a simple and interpretable management tool. This system performs analysis based on two elements: first, it consolidates and visualizes mutual and nontrivial relationships between information features extracted from tickets using multicomponent analysis, bootstrap resampling and ML domain description. Second, it presents statistically relevant relationships in color-coded tables that provide food waste-related recommendations to restaurant managers.FindingsThe study identified relationships between products and customer sales in specific months. Other ticket elements have been related, such as products with days, hours or functional areas and products with products (cross-selling). Big data (BD) technology helped analyze restaurant tickets and obtain information on product sales behavior.Research limitations/implicationsThis study addresses food waste in restaurants using BD and unsupervised ML models. Despite limitations in ticket information and lack of product detail, it opens up research opportunities in relationship analysis, cross-selling, productivity and deep learning applications.Originality/valueThe value and originality of this work lie in the application of BD and unsupervised ML technologies to analyze restaurant tickets and obtain information on product sales behavior. Better sales projection can adjust product purchases to customer demand, reducing food waste and optimizing profits.

Intra and inter-regional functional connectivity of the human brain due to Task-Evoked fMRI Data classification through CNN & LSTM

Background and purposeOlfaction is an early marker of neurodegenerative disease. Standard olfactory function is essential due to the importance of olfaction in human life. The psychophysical evaluation assesses the olfactory function commonly. It is patient-reported, and results rely on the patient's answers and collaboration. However, methodological difficulties attributed to the psychophysical evaluation of olfactory-related cerebral areas led to limited assessment of olfactory function in the human brain. Materials and MethodsThe current study utilized clustering approaches to assess olfactory function in fMRI data and used brain activity to parcellate the brain with homogeneous properties. Deep neural network architecture based on ResNet convolutional neural networks (CNN) and Long Short-Term Model (LSTM) designed to classify healthy with olfactory disorders subjects. ResultsThe fMRI result obtained by k-means unsupervised machine learning model was within the expected outcome and similar to those found with the conn toolbox in detecting active areas. There was no significant difference between the means of subjects and every subject. Proposing a CRNN deep learning model to classify fMRI data in two different healthy and with olfactory disorders groups leads to an accuracy score of 97 %. ConclusionsThe K-means unsupervised algorithm can detect the active regions in the brain and analyze olfactory function. Classification results prove the CNN-LSTM architecture using ResNet provides the best accuracy score in olfactory fMRI data. It is the first attempt conducted on olfactory fMRI data in detail until now.

Using unsupervised machine learning models to drive groundwater chemistry and associated health risks in Indo-Bangla Sundarban region

Groundwater is an essential resource in the Sundarban regions of India and Bangladesh, but its quality is deteriorating due to anthropogenic impacts. However, the integrated factors affecting groundwater chemistry, source distribution, and health risk are poorly understood along the Indo-Bangla coastal border. The goal of this study is to assess groundwater chemistry, associated driving factors, source contributions, and potential non-carcinogenic health risks (PN-CHR) using unsupervised machine learning models such as a self-organizing map (SOM), positive matrix factorization (PMF), ion ratios, and Monte Carlo simulation. For the Sundarban part of Bangladesh, the SOM clustering approach yielded six clusters, while it yielded five for the Indian Sundarbans. The SOM results showed high correlations among Ca2+, Mg2+, and K+, indicating a common origin. In the Bangladesh Sundarbans, mixed water predominated in all clusters except for cluster 3, whereas in the Indian Sundarbans, Cl−-Na+ and mixed water dominated in clusters 1 and 2, and both water types dominated the remaining clusters. Coupling of SOM, PMF, and ionic ratios identified rock weathering as a driving factor for groundwater chemistry. Clusters 1 and 3 were found to be influenced by mineral dissolution and geogenic inputs (overall contribution of 47.7%), while agricultural and industrial effluents dominated clusters 4 and 5 (contribution of 52.7%) in the Bangladesh Sundarbans. Industrial effluents and agricultural activities were associated with clusters 3, 4, and 5 (contributions of 29.5% and 25.4%, respectively) and geogenic sources (contributions of 23 and 22.1% in clusters 1 and 2) in Indian Sundarbans. The probabilistic health risk assessment showed that NO3− poses a higher PN-CHR risk to human health than F− and As, and that potential risk to children is more evident in the Bangladesh Sundarban area than in the Indian Sundarbans. Local authorities must take urgent action to control NO3− emissions in the Indo-Bangla Sundarbans region.

A LDA-Based Social Media Data Mining Framework for Plastic Circular Economy

The mass production of plastic waste has caused an urgent worldwide public health crisis. Although government policies and industrial innovation are the driving forces to meet this challenge, trying to understand public attitudes may improve the efficiency of this process. Social media has become the main ways for the public to obtain information and express opinions and feelings. This motivated us to mine the perceptions and behavioral responses towards plastic usage using social media data. In this paper, we proposed a framework for data collection and analysis based on mainstream media in the UK to obtain public opinions on plastics. An unsupervised machine learning model based on Latent Dirichlet Allocation (LDA) has been employed to analyse and cluster the topics to deal with the lack of annotation of the data contents. An additional dictionary method was then proposed to evaluate the sentiment of the comments. The framework also provides tools to visualise the model and results to stimulate insightful understandings. We validated the framework's effectiveness by applying it to analyse three mainstream social media, where 6 first-level topic categories and 13 second-level topic categories from the comment texts related to plastics have been identified. The results show that public sentiment towards plastic products is generally stable. The spatiotemporal distribution of each topic's sentiment is highly correlated with the number of occurrences.

Explainable representation learning of small quantum states

Unsupervised machine learning models build an internal representation of their training data without the need for explicit human guidance or feature engineering. This learned representation provides insights into which features of the data are relevant for the task at hand. In the context of quantum physics, training models to describe quantum states without human intervention offers a promising approach to gaining insight into how machines represent complex quantum states. The ability to interpret the learned representation may offer a new perspective on non-trivial features of quantum systems and their efficient representation. We train a generative model on two-qubit density matrices generated by a parameterized quantum circuit. In a series of computational experiments, we investigate the learned representation of the model and its internal understanding of the data. We observe that the model learns an interpretable representation which relates the quantum states to their underlying entanglement characteristics. In particular, our results demonstrate that the latent representation of the model is directly correlated with the entanglement measure concurrence. The insights from this study represent proof of concept toward interpretable machine learning of quantum states. Our approach offers insight into how machines learn to represent small-scale quantum systems autonomously.

Visual field loss patterns in dominant and recessive optic atrophy based on OPA1 archetypes

Aims/Purpose: Identification and quantification of visual field loss patterns of dominant optic atrophy forms (ACO2 and AFG3L2 mutations) and recessive optic atrophy form (WFS1 mutation, Wolfram Syndrome, WS) based on machine learning‐obtained archetypes for OPA1‐Dominant Optic Atrophy (DOA).Methods: Five patients affected by hereditary optic neuropathies carrying ACO2 gene mutations, 5 with AFG3L2 mutations, and 20 with WFS1 mutations were enrolled. Visual field (VF) tests performed by SITA standard 30–2 or 24–2 Humphrey VF analyzer (Carl Zeiss Meditec, Dublin, CA, USA) were collected for these patients. The VFs of these rare mutations (30% WFS1, 10% ACO2 and 3% AFG3L2 [1]) were decomposed and quantified according to the archetypes (AT) obtained in the unsupervised machine learning model developed for the OPA1 mutation.Results: The archetype analysis (AA)‐model developed for the DOA‐OPA1 mutation is composed of twelve ATs, with the central, the ceco‐central, and the para‐central as the most significant ones. For WS patients, the VF‐decomposition was mostly made up of ATs resembling more severe abnormalities, while for the ACO2 and AFG3L2‐affected patients the most important ATs were similar to the OPA1 cases.Conclusions: The developed AA‐model based on OPA1‐DOA affected patients enabled the decomposition of VFs of more rare optic atrophy mutations. The association of weights to the different ATs for patient‐specific VFs showed to be a possible option to identify and quantify the most important patterns of visual field loss in such particular cases of autosomal dominant or recessive optic neuropathies. A pattern‐specific visual field loss might strengthen the differential diagnosis between OPA1 and other mutations and support future clinical and therapeutic trials.

Machine learning approaches for fault detection in renewable microgrids

This study focuses on investigating and using machine learning (ML) methods to identify faults in renewable microgrids. It highlights the difficulties and intricacies associated with these dynamic energy systems. The examination of real-world data obtained from solar and wind power production, battery storage status, fault signals, and machine learning model performance highlights the complex nature of fault detection techniques in renewable microgrids. An analysis of data on renewable energy production demonstrates oscillations in the outputs of solar and wind power, highlighting differences of about 5-10% across certain time periods, thereby illustrating the intermittent characteristics of renewable energy sources. Simultaneously, the energy stored in batteries inside the microgrid shows a progressive decrease of about 3-5% in stored energy levels across time intervals, indicating possible consequences for the stability of the system. The fault detection signals display erratic patterns, which emphasize the intricacies involved in finding and categorizing issues inside the system. The assessment of machine learning models, which includes both supervised and unsupervised learning methods, reveals many performance measures. Supervised models provide greater accuracy rates, often ranging from 85% to 90%. However, they are prone to occasional misclassifications. In contrast, unsupervised models provide a moderate level of accuracy, often ranging from 75% to 80%. They exhibit flexibility in detecting faults, but their precision is limited. The study highlights the need of using a combination of supervised and unsupervised machine learning models to improve the accuracy of fault detection in renewable microgrids. These results provide valuable understanding of the intricacies and difficulties of fault detection procedures, which may lead to further progress in improving the dependability and durability of renewable microgrid systems.

Unsupervised machine learning for disease prediction: a comparative performance analysis using multiple datasets

PurposeDisease risk prediction poses a significant and growing challenge in the medical field. While researchers have increasingly utilised machine learning (ML) algorithms to tackle this issue, supervised ML methods remain dominant. However, there is a rising interest in unsupervised techniques, especially in situations where data labels might be missing — as seen with undiagnosed or rare diseases. This study delves into comparing unsupervised ML models for disease prediction.MethodsThis study evaluated the efficacy of seven unsupervised algorithms on 15 datasets, including those of heart failure, diabetes, and breast cancer. It used six performance metrics for this comparison. They are Adjusted Rand Index, Adjusted Mutual Information, Homogeneity, Completeness, V-measure and Silhouette Coefficient.ResultsAmong the seven unsupervised ML methods, the DBSCAN (Density-based spatial clustering of applications with noise) showed the best performance most times (31), followed by the Bayesian Gaussian Mixture (18) and Divisive clustering (15). No single model consistently outshined others across every dataset and metric. The study emphasises the crucial role of model and performance measure selections based on application-specific needs. For example, DBSCAN excels in Homogeneity, Completeness and V-measure metrics. Conversely, the Bayesian Gaussian Mixture is good in the Adjusted R and Index metric. The codes used in this study can be found at https://github.com/haohuilu/unsupervisedml/.ConclusionThis research contributes deeper insights into the unsupervised ML applications in healthcare and encourages further investigations into model selection. Subsequent studies could harness genuine disease records for a more nuanced comparison and evaluation of models.

UNSUPERVISED WINTER WHEAT MAPPING BASED ON MULTI-SPECTRAL AND SYNTHETIC APERTURE RADAR OBSERVATIONS

Abstract. Annual meteorological variations and the impact of climate change in recent years impacted on agricultural production and distribution. Since wheat is a main food resource and the most widely grown crop in the world, it is essential to ensure the sustainability of its production. Therefore, accurate wheat mapping is essential for agricultural production forecasts. Multi-spectral satellite image analysis, including supervised machine learning (ML) methods, has been applied to wheat and other crop mapping, but such passive, optical imaging approaches are strongly influenced by weather conditions and cloud cover, whilst the supervised ML algorithms are highly reliant on manual labelling and ground control data. To avoid the limitation of weather and cloud, this research integrates Sentinel-1 Synthetic Aperture Radar (SAR) data with Sentinel-2 multi-spectral image products to achieve more reliable and accurate winter wheat mapping. Normalised Difference Vegetation Index (NDVI) retrieved from multi-spectral imagery and Sentinel-1’s dual-polarisation radar signals, VV, VH, acquired in different time series, are used as key inputs to an unsupervised ML model based on Dynamic Time Warping (DTW) and hierarchical clustering to prevent time-consuming manual labelling. The chosen study area lies in Norfolk, UK. The result of winter wheat classification with NDVI time series data in this study reaches 72% accuracy, but the improved classification integrating NDVI, VH, VV and VH/VV values achieves 86% accuracy. Future research will focus on optimizing the ML model with multi data integration in addition additional research sites will test more complicated scenarios and multiple crop classification.

Insider Threat Detection Based on Deep Clustering of Multi-Source Behavioral Events

With the continuous advancement of enterprise digitization, insider threats have become one of the primary cybersecurity concerns for organizations. Therefore, it is of great significance to develop an effective insider threat detection mechanism to ensure the security of enterprises. Most methods rely on artificial feature engineering and input the extracted user behavior features into a clustering-based unsupervised machine learning model for insider threat detection. However, feature extraction is independent of clustering-based unsupervised machine learning. As a result, user behavior features are not the most appropriate for clustering-based unsupervised machine learning, and thus, they reduce the insider threat detection accuracy. This paper proposes an insider threat detection method based on the deep clustering of multi-source behavioral events. On the one hand, the proposed method constructs an end-to-end deep clustering network and automatically learns the user behavior feature expression from multi-source behavioral event sequences. On the other hand, a deep clustering objective function is presented to jointly optimize the learning of feature representations and the clustering task for insider threat detection. This optimization can adjust the optimal user behavior features for the clustering model to improve the insider threat detection accuracy. The experimental results show that the proposed end-to-end insider threat detection model can accurately identify insider threats based on abnormal multi-source user behaviors in enterprise networks.

Unsupervised Detection of Covariate Shift Due to Changes in EEG Headset Position: Towards an Effective Out-of-Lab Use of Passive Brain–Computer Interface

In the field of passive Brain–computer Interfaces (BCI), the need to develop systems that require rapid setup, suitable for use outside of laboratories is a fundamental challenge, especially now, that the market is flooded with novel EEG headsets with a good quality. However, the lack of control in operational conditions can compromise the performance of the machine learning model behind the BCI system. First, this study focuses on evaluating the performance loss of the BCI system, induced by a different positioning of the EEG headset (and of course sensors), so generating a variation in the control features used to calibrate the machine learning algorithm. This phenomenon is called covariate shift. Detecting covariate shift occurrences in advance allows for preventive measures, such as informing the user to adjust the position of the headset or applying specific corrections in new coming data. We used in this study an unsupervised Machine Learning model, the Isolation Forest, to detect covariate shift occurrence in new coming data. We tested the method on two different datasets, one in a controlled setting (9 participants), and the other in a more realistic setting (10 participants). In the controlled dataset, we simulated the movement of the EEG cap using different channel and reference configurations. For each test configuration, we selected a set of electrodes near the control electrodes. Regarding the realistic dataset, we aimed to simulate the use of the cap outside the laboratory, mimicking the removal and repositioning of the cap by a non-expert user. In both datasets, we recorded multiple test sessions for each configuration while executing a set of Workload tasks. The results obtained using the Isolation Forest model allowed the identification of covariate shift in the data, even with a 15-s recording sample. Moreover, the results showed a strong and significant negative correlation between the percentage of covariate shift detected by the method, and the accuracy of the passive BCI system (p-value < 0.01). This novel approach opens new perspectives for developing more robust and flexible BCI systems, with the potential to move these technologies towards out-of-the-lab use, without the need for supervision for use by a non-expert user.