Automated explainable machine learning for squeeze-casting: automatic relationship analysis

Purpose: Maternal mortality remained to persist in many developing countries. India being the most populous developing country has several cultural differences and beliefs on health systems and may have disparities in receiving proper maternal health care. High data availability and under-utilization of data centric decision making is a key reason for ineffective performance of health systems. Enacting machine learning on such data to aid in sub-divisional policy formulation to address area level problems will eradicate major disparities in recipients of health services. Methods: Glass box machine learning models are trained on the data to obtain importance of features in defining the maternal mortality of a district. Furthermore, black box machine learning models are trained with hyper-parameter tuning and best model is chosen to perform explainable machine learning to generate explanations for each district prediction. A hybrid explainable machine learning approach is proposed on black-box machine learning models where Shapley Additive Explanation and Local Interpretable Model-agnostic Explanations are combined to generate final explanations. Results: There may be several differences even among nearby districts. Health Management Information System data is analyzed with help of Machine Learning techniques and Explainable Machine Learning techniques are used on the trained models to evaluate the contributing factors for each district. Conclusion: The factors that are specific to each district can help in formulating region specific health policies that minimize the disparities of progress of preventing maternal mortality over the districts of India. The paper has highlighted the advantages of using explainable machine learning in extracting complicated patterns of the data.

Most questions in the Fish4Knowledge project are related to the ability to count fish, fish species or events using video analysis Chap. 2. Automatic video analysis however brings uncertainty due to False Positive/False Negative classifications, which makes determining the counts based on automatic video analysis difficult. Automatic video analysis software is often able to express a measure of uncertainty for a single detection/recognition using a similarity score, indicating how certain the software is about a single decision. Logistic Regression allows us to combine these similarity scores to compute an estimated count, outperforming the estimates of the underlying machine learning methods used for the original video analysis. We show this works both for the two-class and multi-class problem. Furthermore, we identify potential pitfalls where ensuring a correct sampling procedure is essential. The error in the estimated species counts using Logistic Regression was on average around 15 on a set of 11,585 fish images, while the machine methods had an average error of 800. The key to understanding this huge improvement in accuracy is that we use Logistic Regression over the individual classifications to estimate the size of the whole population, rather than the species classification of any individual observation.

To better understand degradation in electrochemical converters and helping to correlate certain phenomena with specific operating conditions, machine learning (ML) methods are increasingly being applied. Success has already been achieved in the field of degradation analysis and prediction of capacity of lithium ion batteries1, for instance. In terms of Solid Oxide Cell (SOC) stacks ML methods have been applied mainly with the aim of identification of faulty operation modes and degradation related fault diagnosis2. ML approaches usually require a considerable amount of real training data, when used for forecasting models. A data consolidation and curation strategy was developed with the aim of processing the historic long-term test bench data of SOCs collected by Forschungszentrum Jülich over the past years. In comparison to other datasets developed in this field3, the one presented in this work contains SOC stack tests in fuel cell operation with significantly longer operating times under load. A compilation of the sample experiments and the consolidation into a hierarchical data format are presented. Further, an essential part of the strategy is the automatic curation and analysis of electrochemical impedance spectroscopy (EIS) measurements, using a specifically developed procedure in Python. The varying quality of measurements from past years, as well as recurring artefacts such as parasitic inductances, can be addressed in this way. Additional distribution of relaxation times (DRT) deconvolutions and equivalent circuit modelling (ECM) are performed, as part of the procedure to automatically retrieve feature values from measurements (cf. Fig. 1). The novel dataset, which to the authors’ knowledge includes some of the longest SOC stack tests available, serves as the basis for several evaluations. In addition to classification and clustering work to derive degradation patterns, in particular based on the EIS data, another focus is on the development of forecasting models. The current work is primarily concerned with long short-term memory (LSTM), as well as regression models that make use of both the time series data and the characterisation measurements, such as EIS. Acknowledgement The authors would like to thank their colleagues at Forschungszentrum Jülich GmbH for their great support and the Helmholtz Society as well as the German Federal Ministry of Education and Research for financing these activities as part of the WirLebenSOFC project (03SF0622B). References 1: Jones, P.K., Stimming, U. & Lee, A.A. Impedance-based forecasting of lithium-ion battery performance amid uneven usage. Nature Communications 13, 4806 (2022).2: B. Yang et al. Solid oxide fuel cell systems fault diagnosis: Critical summarization, classification, and perspectives. Journal of Energy Storage 34, 102153 (2021).3: A.K. Padinjarethil, S. Pollok & A. Hagen. Degradation studies using machine learning on novel solid oxide cell database. Fuel Cells 21, 566–576 (2021). Figure caption: Fig.1: Flow diagram of EIS data curation pipeline and curation results for example EIS measurement. Figure 1

The first objective of this study was to apply computer vision and machine learning techniques to quantify the effects of haircare treatments on hair assembly and to identify correctly whether unknown tresses were treated or not. The second objective was to explore and compare the performance of human assessment with that obtained from artificial intelligence (AI) algorithms. Machine learning was applied to a data set of hair tress images (virgin and bleached), both untreated and treated with a shampoo and conditioner set, aimed at increasing hair volume whilst improving alignment and reducing the flyway of the hair. The automatic quantification of the following hair image features was conducted: local and global hair volumes and hair alignment. These features were assessed at three time points: t0 (no treatment), t1 (two treatments) and t2 (three treatments). Classifier tests were applied to test the accuracy of the machine learning. A sensory test (paired comparison of t0 vs t2 ) and an online front image-based survey (paired comparison of t0 vs t1 , t1 vs t2 , t0 vs t2 ) were conducted to compare human assessment with that of the algorithms. The automatic image analysis identified changes to hair volume and alignment which enabled the successful application of the classification tests, especially when the hair images were grouped into untreated and treated groups. The human assessment of hair presented in pairs confirmed the automatic image analysis. The image assessment for both virgin hair and bleached only partially agreed with the analysis of the subset of images used in the online survey. One hypothesis is that treatments changed somewhat the shape of the hair tress, with the effect being more pronounced in bleached hair. This made human assessment of flat images more challenging than when viewed directly in 3D. Overall, the bleached hair exhibited effects of higher magnitude than the virgin hair. This study illustrated the capacity of artificial intelligence for hair image detection and classification, and for image analysis of hair assembly features following treatments. The human assessment partially confirmed the image analysis and highlighted the challenges imposed by the presentation mode.

The evaluation and prediction of volcanoes activities and associated risks is still a timely and open issue. The amount of volcano-seismic data acquired by recent monitoring stations is huge (e.g., several years of continuous recordings), thereby making machine learning absolutely necessary for their automatic analysis. The transient nature of the volcano-seismic signatures of interest further enforces the need of automatic detection and classification of such events. In this paper, we present a novel architecture for automatic classification of volcano-seismic events based on a comprehensive signal representation with a large feature set. To the best of our knowledge this is one of the first attempts to automatize the classification task of these signals. The proposed approach relies on supervised machine learning techniques to build a prediction model.

Management of phosphate mine waste rock piles (PWRPs) is a critical challenge in the mining industry, particularly in regions like Morocco, which holds the world’s largest phosphate reserves. To this end, there is a need for an approach that focuses on real-time monitoring of waste rock heterogeneity, enabling more efficient resource recovery and environmental management. This study proposes a novel, multi-scale approach that integrates hyperspectral imaging, field spectroscopy, and explainable machine learning (XML) to characterize and map the mineralogical diversity of PWRPs at the Benguerir mine.  A total of 103 samples were collected from waste rock piles across an area of approximately 60 km², representing the full spectrum of mineralogical variability. Handheld X-ray fluorescence (XRF) analysis was conducted on the all the samples and revealed the dominance of SiO₂ (29.51 wt% ± 12.42), CaO (30.16 wt% ± 10.17), and P₂O₅ (7.23 wt% ± 4.21). These XRF analyses indicated the presence of silicate, carbonate, and phosphate-bearing materials. These findings were complemented by both PRISMA hyperspectral imaging, which captured spectral data across the visible to shortwave infrared (VSWIR) range. precise calibration and validation of the remote sensing outputs were conducted using field spectroscopy using the ASD FieldSpec 4 spectroradiometer.To address the complexity of the spectral data, we developed an explainable machine learning framework based on SHapley Additive exPlanations (SHAP) and Convolutional Neural Networks (CNN). This framework not only improved classification accuracy (achieving 0.92 overall accuracy) but also provided interpretable insights into the spectral features driving mineral identification. Our results showed that the used model successfully differentiated four main waste rock categories: carbonate-rich, phosphate-rich, clay-dominated, and siliceous materials. The resulting maps offer a practical tool for real-time waste management and resource recovery. For instance, carbonate-rich materials, characterized by high CaO content, can be identified or used for construction applications, while phosphate-rich zones, with elevated P₂O₅ levels, can be flagged for potential recovery and further processing. This targeted approach ensures that waste materials are repurposed efficiently, aligning with circular economy principles. The study highlights the potential for automated, spectroscopy-based monitoring systems to support sustainable mining practices. Overall, this study demonstrates the power of combining cutting-edge remote sensing technologies with explainable machine learning to address the challenges of phosphate waste rock characterization. The methodology provides a scalable, cost-effective solution for mining operations worldwide, with significant implications for environmental sustainability, resource efficiency, and circular economy initiatives.Keywords: Phosphate mine waste, Hyperspectral imaging, Field spectroscopy, Explainable machine learning (XML), Sustainable mining.

It is well known that the interpretation of cardiotocographic (CTG) signals is still subjective and prone to misinterpretations; as a consequence, there has been an increase of cesarean sections, often not necessary, and initial expectations of significantly reducing perinatal morbidity and mortality have been unattended. Nevertheless, in developed countries, CTG is still the most widely employed prenatal technique for monitoring fetal health; and in many countries it represents a medical report with legal value. To overcome the drawbacks of the visual interpretation, many computerized systems for automatic or semi-automatic analysis of CTG have been developed in the last years. Recently, in order to support the diagnosis process, and to increase the predictive capability of these systems, also other techniques such as artificial neural network, deep learning and machine learning have been introduced. In previous works of the authors, software for automatic CTG analysis has been developed and described in detail. Now, by employing a dataset of features extracted from CTG signals with that software, to enhance its performances, different algorithms, such as J48, Adaboosting, Random Forests and Gradient Boosted Tree, have been tested to predict whether a birth would be a caesarean section or a vaginal delivery. The RF algorithm showed the best performance, since it reached the highest accuracy (87.6%), precision (87.9%) and AUCROC (93.0%). These preliminary results are very satisfying and encouraging; they confirm that to enrich the CTG analysis software with this methodology can help to significantly improve CTG classification.

BackgroundDespite the high prevalence and significant burden of depression, underdiagnosis remains a persistent challenge. Automatic speech analysis (ASA) has emerged as a promising method for depression assessment. However, a comprehensive quantitative synthesis evaluating its diagnostic accuracy is still lacking.ObjectiveThis systematic review and meta-analysis aimed to assess the diagnostic performance of ASA in detecting depression, considering both machine learning and deep learning approaches.MethodsWe conducted a systematic search across 8 databases, including MEDLINE, PsycInfo, Embase, CINAHL, IEEE Xplore, ACM Digital Library, Scopus, and Google Scholar from January 2013 to April 1, 2025. We included studies published in English that evaluated the accuracy of ASA for detecting depression, and reported performance metrics such as accuracy, sensitivity, specificity, precision, or confusion matrices. Study quality was assessed using a modified version of the Quality Assessment of Studies of Diagnostic Accuracy-Revised. A 3-level meta-analysis was performed to estimate the pooled highest and lowest accuracy, sensitivity, specificity, and precision. Meta-regressions and subgroup analyses were performed to explore heterogeneity across various factors, including type of publication, artificial intelligence algorithms, speech features, speech-eliciting tasks, ground truth assessment, validation approach, dataset, dataset language, participants’ mean age, and sample size.ResultsOf the 1345 records identified, 105 studies met the inclusion criteria. The pooled mean of the highest accuracy, sensitivity, specificity, and precision were 0.81 (95% CI 0.79 to 0.83), 0.84 (95% CI 0.81 to 0.86), 0.83 (95% CI 0.79 to 0.86), and 0.81 (95% CI 0.77 to 0.84), respectively, whereas the pooled mean of the lowest accuracy, sensitivity, specificity, and precision were 0.66 (95% CI 0.63 to 0.69), 0.63 (95% CI 0.58 to 0.68), 0.60 (95% CI 0.55 to 0.66), and 0.64 (95% CI 0.58 to 0.70), respectively.ConclusionsASA shows promise as a method for detecting depression, though its readiness for clinical application as a standalone tool remains limited. At present, it should be regarded as a complementary method, with potential applications across diverse contexts. Further high-quality, peer-reviewed studies are needed to support the development of robust, generalizable models and to advance this emerging field.Trial RegistrationPROSPERO CRD42023444431; https://www.crd.york.ac.uk/PROSPERO/view/CRD42023444431

Drilling dynamics measurement of drilling motors and its application in recognition of motor operation states through machine learning

Spatial-temporal patterns and drivers of coastal water quality dynamics: Insights from explainable machine learning and PLS-SEM analysis in Xiamen Bay, China.

Multispectral imaging has become a useful means of sensing in increasing number of fields, including precision agriculture due to its many advantages such as being non-invasive and extracting additional information than the visible band. Combined with machine learning, such systems can further extract complex relationships among extracted spectral information to facilitate intelligent agriculture. This paper focuses on making the process of multispectral data analysis automatic for classifying virus infection in plants, specially cassava plants infected with brown streak virus and tomato plants infected with mottle virus. Multispectral images are sampled from cassava and tomato plants with a designed portable device and then passed on to the proposed automatic analysis process. Conventional methods often depend on time-consuming manual processes of cropping valid patches from sampled leaf images and then averaging the patches to obtain the spectral reflectance signatures of the leaves, prior to the classification stage. The developed automatic process can automatically extract valid leaf pixels from scanned leaf images and subsequent spectral information and integrate with the classification method for optimal detection of infected plants. It has not only reduced processing time and errors significantly but also make the entire process more optimal. Extensive experiments on cassava brown streak virus and tomato mottle virus have been conducted and results demonstrated the intended advantages of the developed process. Support vector machines with RBF kernel have been shown to perform well for the classification of uninfected, and infected classes. Experiments show that the application can offer less time-consuming automatic analysis of the captured data.KeywordsMultispectral data analysisClassificationPrecision agricultureVirus detection

We propose an approach for complex software analysis based on visualization. Our work is motivated by the fact that in spite of years of research and practice, software development and maintenance are still time and resource consuming, and high-risk activities. The most important reason in our opinion is the complexity of many phenomena related to software, such as its evolution and its reliability. In fact, there is very little theory explaining them. Today, we have a unique opportunity to empirically study these phenomena, thanks to large sets of software data available through open-source programs and open repositories. Automatic analysis techniques, such as statistics and machine learning, are usually limited when studying phenomena with unknown or poorly-understood influence factors. We claim that hybrid techniques that combine automatic analysis with human expertise through visualization are excellent alternatives to them. In this paper, we propose a visualization framework that supports quality analysis of large-scale software systems. We circumvent the problem of size by exploiting perception capabilities of the human visual system.

IHC-Net: A fully convolutional neural network for automated nuclear segmentation and ensemble classification for Allred scoring in breast pathology

Engineering drawings are commonly used across different industries such as oil and gas, mechanical engineering and others. Digitising these drawings is becoming increasingly important. This is mainly due to the legacy of drawings and documents that may provide rich source of information for industries. Analysing these drawings often requires applying a set of digital image processing methods to detect and classify symbols and other components. Despite the recent significant advances in image processing, and in particular in deep neural networks, automatic analysis and processing of these engineering drawings is still far from being complete. This paper presents a general framework for complex engineering drawing digitisation. A thorough and critical review of relevant literature, methods and algorithms in machine learning and machine vision is presented. Real-life industrial scenario on how to contextualise the digitised information from specific type of these drawings, namely piping and instrumentation diagrams, is discussed in details. A discussion of how new trends on machine vision such as deep learning could be applied to this domain is presented with conclusions and suggestions for future research directions.

Circulating blood cell clusters (CCCs) containing red blood cells (RBCs), white blood cells (WBCs), and platelets are significant biomarkers linked to pathological conditions like thrombosis, infection, and inflammation. Flow cytometry, paired with fluorescence staining, is commonly used to analyze these cell clusters, revealing cell morphology and protein profiles. While computational approaches based on machine learning have advanced the automatic analysis of single-cell flow cytometry images, there is a lack of effort to build tools to automatically analyze images containing CCCs. Unlike single cells, cell clusters exhibit irregular shapes and sizes. In addition, these cell clusters often consist of heterogeneous cell types, which require multi-channel staining to identify the specific cell types within the clusters. To address these challenges, we introduce a new computational framework for analyzing CCC images and identifying cell types within clusters. Our framework uses a two-step analysis strategy. First, it categorizes images into cell cluster and non-cluster groups by fine-tuning the You Only Look Once (YOLOv11) model, which outperforms traditional convolutional neural networks (CNNs), such as Vision Transformers (ViT). Then, it identifies cell types by overlaying cluster contours with regions from multi-channel fluorescence stains, thereby minimizing the impact of cell debris and staining artifacts. This approach achieved over 95% accuracy in both cluster classification and cell phenotype identification. In summary, our automated framework effectively analyzes CCC images from flow cytometry, leveraging both bright-field and fluorescence data. Initially tested on blood cells, it holds potential for broader applications, such as analyzing immune and tumor cell clusters, supporting cellular research across various diseases.