This work explores integrating large language models (LLMs) with the adaptive-network-based fuzzy inference system (ANFIS) to enhance interpretability and usability in decision-making processes. ANFIS generates transparent and interpretable fuzzy rules, while LLMs complement these by providing concise, context-aware textual explanations. By combining ANFIS’s data-driven capabilities with the semantic understanding of LLMs, the framework aims to clarify AI outputs and support error identification in the knowledge base. The pipeline implements a human-in-the-loop strategy to engage domain experts in enhancing prompts, verifying explanations, and aligning outputs with expert standards. The methodology was assessed in a medical setting, particularly for predicting epilepsy seizures using EEG data. This study illustrates how the proposed pipeline bridges AI models and real-world applications, providing transparent insights into decision-making processes. It lays the groundwork for creating more interactive, accurate, explainable, and user-friendly tools for predictive analytics, particularly in critical fields like healthcare.

We apply an evolving granular-computing modeling approach, called evolving Optimal Granular System (eOGS), to bipolar mood disorder (BD) diagnosis based on speech data streams. The eOGS online learning algorithm reveals information granules in the flow and design the structure and parameters of a granular rule-based model with a certain degree of interpretability based on acoustic attributes obtained from phone calls made over 7 months to the Psychiatry department of a hospital. A multi-objective programming problem that trades-off information specificity, model compactness, and numerical and granular error indices is presented. Spectral and prosodic attributes are ranked and selected based on a hybrid Pearson-Spearman correlation coefficient . Low attribute-class correlation, ranging from 0.03 to 0.07, is observed, as well as high class overlap, which is typical in the psychiatric field. eOGS models for BD recognition overcome alternative computational-intelligence models, namely, Dynamic Evolving Neural-Fuzzy Inference System (DENFIS) and Fuzzy-set-Based evolving Modeling (FBeM-Gauss), by a small margin in both best and average cases; followed by eXtended Takagi-Sugeno (xTS) and evolving Takagi Sugeno (eTS) types of models. The proposed eOGS model using only 8 of the original acoustic attributes, and about 15 ‘If-Then’ inference rules, has exhibited the best root mean square error , 0.1361, and 91.8% accuracy in sharp BD class estimates. Granules associated to linguistic labels and a granular input-output map offer human understandability with relation to the inherent process of generating class estimates. Linguistically readable eOGS rules may assist physicians in explaining symptoms and making a diagnosis.

This paper introduces a novel pipeline for estimating systolic and diastolic blood pressure using remote photoplethysmographic (rPPG) signals derived from video recordings of subjects’ faces. The pipeline consists of three main stages: rPPG signal extraction, denoising to transform the rPPG signal into a PPG-like waveform, and blood pressure estimation. This approach directly addresses the current lack of datasets that simultaneously include video, rPPG, and blood pressure data. To overcome this, the proposed pipeline leverages the extensive availability of PPG-based blood pressure estimation techniques, in combination with state-of-the-art algorithms for rPPG extraction, enabling the generation of reliable PPG-like signals from video input. To validate the pipeline, we conducted comparative analyses with state-of-the-art methods at each stage and collected a dedicated dataset through controlled laboratory experimentation. The results demonstrate that the proposed solution effectively captures blood pressure information, achieving a mean error of 9.2 ± 11.3 mmHg for systolic and 8.6 ± 9.1 mmHg for diastolic blood pressure. Moreover, the denoised rPPG signals show a strong correlation with conventional PPG signals, supporting the reliability of the transformation process. This non-invasive and contactless method offers considerable potential for long-term blood pressure monitoring, particularly in Ambient Assisted Living (AAL) systems, where unobtrusive and continuous health monitoring is essential.

Predicting whether a newly submitted bug will be resolved quickly or slowly is a crucial aspect of the bug triage process, as it enables project managers to estimate software maintenance efforts and manage development workflows more effectively. This paper proposes a deep learning approach for classifying bug reports into two categories— FAST or SLOW —based on their expected fixing time. The method leverages a feature set composed of the bug description and reporter comments and adopts a transfer learning strategy using pre-trained Large Language Models (LLMs). The problem is framed as a supervised text classification task, where LLMs exploit their ability to learn rich contextual representations of language. We introduce a novel classification workflow that guides the LLM through a structured prompt, combining two design patterns: the persona pattern to contextualize the task and the input semantic pattern to organize textual information. The workflow relies on zero-shot learning to assess whether the intrinsic knowledge embedded in the LLMs is sufficient for this prediction task. We conducted a comprehensive evaluation of three state-of-the-art LLMs across multiple real-world datasets sourced from Bugzilla, encompassing a diverse range of software projects. The experimental results demonstrate that the proposed method is effective in accurately identifying fast-resolving bugs. Among the evaluated models, LLaMA3-8B consistently delivered superior performance. Additionally, the absence of statistically significant performance variations across datasets highlights the generalizability of the approach. Notably, the LLMs maintained strong performance even on small and imbalanced datasets, underscoring their robustness and practical applicability in real-world, data-scarce scenarios.

We propose a method to perform regression on partially labeled data, which is based on SSFCM (Semi-Supervised Fuzzy C-Means), an algorithm for semi-supervised classification based on fuzzy clustering. The proposed method, called SSFCM-R, precedes the application of SSFCM with a relabeling module based on target discretization. After the application of SSFCM, regression is carried out according to one out of two possible schemes: (i) the output corresponds to the label of the closest cluster; (ii) the output is a linear combination of the cluster labels weighted by the membership degree of the input. Some experiments on synthetic data are reported to compare both approaches.

The current advances in wearable sensors show the shining future of socially implemented Internet-of-Medical-Things (IoMT) devices (e.g., smartwatches). However, the recent machine learning approaches cannot be applied well in these devices, because almost all the processing in the IoMT devices is now being performed in classic forms (mainly as centralized computing) or based on cloud services. This topical collection has tried to extend our knowledge about how to apply collaborative learning to IoMT considering social edge/fog nodes’ facilities.

Semi-supervised learning has gained great interest because of its ability to combine unlabeled data with – potentially few – labeled observations in a training process. However, in some application contexts, one can question whether all available labels are equally valid. For example, in the context of bipolar disorder (BD) remote monitoring, a common practice is to extrapolate the psychiatrist’s assessment onto some fixed time window surrounding the visit, the so-called ground truth period. In consequence, all data from this period are labeled with the same category. Such an approach may potentially result in misguided supervision affecting the model’s performance. In this paper, we consider the problem of label uncertainty, assuming that the labels are crisp, but they may be assigned to particular observations with varying confidence. We propose a novel method called Confidence Path Regularization (CPR) that incorporates this uncertainty into the fuzzy c-means semi-supervised learning. The proposed CPR approach is a novel method for automatic, data-driven handling of label uncertainty. We achieve it by estimating the confidence factor for each labeled observation. In addition, CPR allows for the exploration of potential class-specific patterns in the adjusted confidence. The proposed method is illustrated with experiments on partially labeled data about speech characteristics collected from smartphone application for BD monitoring. In this particular applied scenario, we also use additional contextual data to improve the construction of confidence paths. It is shown that the proposed CPR approach enables to reflect the varying confidence in labels as compared with the nominal approach which assigns the majority of observations to the same class associated with relevant ground truth period

Intelligent systems for the medical domain often require processing data streams that evolve over time and are only partially labeled. At the same time, the need for explanations is of utmost importance not only due to various regulations, but also to increase trust among systems’ users. In this work, an online data-driven learning method with focus on the explainability of evolving models equipped with incremental semi-supervised learning algorithms is considered. The proposed method combines: (i) the Dynamic Incremental Semi-Supervised Fuzzy C-Means (DISSFCM) algorithm to incrementally classify subsets of data; with (ii) Linguistic Summarization, which provides explanations of the classification results in terms of short sentences in a natural language. The approach has been illustrated for streaming data collected from voice calls of patients affected by Bipolar Disorder. The results show the effectiveness of the proposed method in classifying instances belonging to healthy and affective states, and explaining the approximate reasoning behind the classification of new acoustic data related to patients.

The healthcare domain has undergone a huge transformation thanks to the availability of new technologies. In particular, health monitoring systems have entered everyday life without interfering with the daily routine. Mobile phones are increasingly used as health monitoring systems by means of ad-hoc applications. In this work, we propose a mobile app for contactless monitoring of vital signs, such as heart rate and blood oxygen saturation. Differently from the other devices in the literature, it is able to measure vital signs from the analysis of short videos through the use of remote photoplethysmography technology. A client-server architecture has been developed to run the signal and video processing on the server while implementing video acquisition and communication with the user on the smartphone. Experiments have shown the effectiveness of the developed app in accurately measuring vital parameters.

Smartphones enable to collect large data streams about phone calls that, once combined with Computational Intelligence techniques, bring great potential for improving the monitoring of patients with mental illnesses. However, the acoustic data streams recorded in uncontrolled environments are dynamically changing due to various sources of uncertainty. In addition, such acoustic data are usually difficult to interpret by psychiatrists. Within this study, we propose an approach based on Linguistic Summaries with Fuzzy Clustering (LS-FC) aiming at the development of human-consistent and easily interpretable summaries about relations between acoustic data and mental state of a patient affected by Bipolar Disorder, e.g., Most calls in the state of hypomania have low loudness compared to the state of euthymia [T = 1]. To capture the dynamics of acoustic data streams, we apply a dynamic incremental semi-supervised fuzzy clustering that synthesizes data into clusters. These clusters are represented by prototypes which are used for the construction of the membership functions describing linguistic terms e.g., low loudness, and then, linguistic summaries. The main contribution of this paper is the incorporation of information about clusters’ prototypes in the generation of linguistic summaries. The primary goal of this research is explainability. The semi-supervised learning algorithm is used mainly for deriving clusters and building improved linguistic summaries. Numerical results indicate that linguistic summaries provide intuitive and clear information about voice features in a patient’s affective state and they are consistent with clinical observation. In particular, during most calls in hypomania/mania both the quality of the patient’s voice and the dynamics of change in the spectrum signal reflected in spectral flux are low compared to euthymia. The proposed approach enables to summarize large data streams into meaningful descriptions that, although relatively simple, offer information granules that are very intuitive for clinicians and are promising to support the smartphone-based monitoring of bipolar disorder patients to inform about the potential change of mental state.

We introduce an approach called PLENARY (exPlaining bLack-box modEls in Natural lAnguage thRough fuzzY linguistic summaries), which is an explainable classifier based on a data-driven predictive model. Neural learning is exploited to derive a predictive model based on two levels of labels associated with the data. Then, model explanations are derived through the popular SHapley Additive exPlanations (SHAP) tool and conveyed in a linguistic form via fuzzy linguistic summaries. The linguistic summarization allows translating the explanations of the model outputs provided by SHAP into statements expressed in natural language. PLENARY accounts for the imprecision related to model outputs by summarizing them into simple linguistic statements and for the imprecision related to the data labeling process by including additional domain knowledge in the form of middle-layer labels. PLENARY is validated on preprocessed speech signals collected from smartphones from patients with bipolar disorder and on publicly available mental health survey data. The experiments confirm that fuzzy linguistic summarization is an effective technique to support meta-analyses of the outputs of AI models. Also, PLENARY improves explainability by aggregating low-level attributes into high-level information granules, and by incorporating vague domain knowledge into a multi-task sequential and compositional multilayer perceptron. SHAP explanations translated into fuzzy linguistic summaries significantly improve understanding of the predictive modelling process and its outputs.

Abstract We study the impact of fuzziness on the behavior of Fuzzy Rule-Based Classifiers (FRBCs) defined by trapezoidal fuzzy sets forming Strong Fuzzy Partitions. In particular, if an FRBC selects the class related to the rule with the highest activation (so-called Winner-Takes-All approach), then fuzziness, as quantified by the slope of the membership functions, has no impact in classifying data in regions of the input space where rules dominate. On the other hand, fuzziness affects the behaviour of the FRBC in regions where the confidence in classification is low. As a consequence, in the context of Explainable Artificial Intelligence, fuzziness is profitable in FRBCs only if classification is accompanied by an explanation of the confidence of the provided outputs.

Voice features from everyday phone conversations are regarded as a sensitive digital marker of mood phases in bipolar disorder. At the same time, although acoustic data collected from smartphones are relatively large, their psychiatric labelling is usually very limited, and there is still a need for intelligent and interpretable approaches to process such multiple data streams with a low percentage of labelling. Furthermore, both acoustic data and psychiatric labels are subject to several sources of uncertainty (e.g., irregular phone usage, background noises, subjectivity in psychiatric evaluation). To cope with these characteristics of an acoustic data stream, this paper introduces an intelligent qualitative and quantitative analysis based on the Dynamic Incremental Semi-Supervised Fuzzy C-Means algorithm (DISSFCM) for supporting bipolar disorder monitoring. The proposed approach is illustrated with real-life data collected from smartphones and psychiatric assessments of a bipolar disorder patient. Analysis of the dynamics of data streams basing on the cluster prototypes from fuzzy semi-supervised learning is a highly novel approach. It is also showed that the DISSFCM algorithm obtains relatively high classification performance (accuracy ranging from 0.66 to 0.76) already with 25% labelling percentage, thanks to the splitting mechanism that is adapting the number of clusters to the structure of data.

Mobile health (mHealth) technologies play a fundamental role in epidemiological situations such as the ongoing outbreak of COVID-19 because they allow citizen to self-monitor their health status while staying at home and being constantly in remote connection with the physicians despite the quarantine. Special care should be given to self-monitoring vital parameters such as blood oxygen saturation (SpO2), whose abnormal values are a warning sign for potential infection by COVID-19. Measurement of SpO2 is commonly made through the pulse oximeter that requires skin contact and hence could be a potential way of spreading contagious infections. For this reason, contact-less solutions for self-monitoring of SpO2 would be beneficial. In this paper we present a mHealth approach to self-monitor SpO2 that does not require any contact device since it is based on video processing. Video frames of the patient's face acquired by a camera are processed in real-time in order to extract the remote photoplethysmography signal useful to derive an estimation of SpO2. Preliminary experimental results show that the SpO2 values obtained by our contact-less solution are consistent with the measurements of a commercial pulse oximeter used as reference device.

Virtual Learning Environments (VLE) offer a wide range of courses and learning supports for students. Such innovative learning platforms generate daily a huge quantity of data, regarding the interactions among the students and the VLE. To analyze these big educational data a new research branch called educational data mining (EDM) has emerged, that puts together computer scientists and pedagogues researchers' expertise. So far, educational data have been studied as stationary data by traditional machine learning methods. Rather, educational data are non-stationary in nature and can be better analyzed as data streams. In this paper we investigate the use of an adaptive fuzzy clustering algorithm called DISSFCM (Dynamic Incremental Semi-Supervised FCM) to process educational data as data streams and predict the students' outcomes to one exam module. Numerical experiments on the Open University Learning Analytics Dataset (OULAD) show the reliability of DISSFCM in creating good classification models of educational data.