Time-series classification with Large Language Models (LLMs) remains challenging because raw continuous signals lack the discrete sequential structure that language models are designed to process. We observe that Symbolic Aggregate approXimation (SAX) naturally produces ordered token sequences structurally compatible with LLM inputs, making the symbolic-LLM connection principled rather than ad hoc. Building on this observation, we propose SAX_HAR-LLM, a dual-stream framework that encodes inertial sensor data as symbolic sequences augmented with lightweight kinematics-informed descriptors, processed jointly by a fine-tuned causal language model for human activity recognition. The framework is evaluated on two benchmark inertial HAR datasets — WISDM and MotionSense — under a subject-independent protocol against symbolic, deep learning, and kernel-based baselines. Ablation experiments provide direct empirical evidence for the central theoretical claim: the fine-tuned LLM benefits from pre-trained sequential priors when processing symbolic time-series representations. Dynamic activities are discriminated from symbolic sequences alone, while kinematics-informed descriptors function as a targeted auxiliary mechanism for classes becomimg symbolically indistinguishable under z-normalization. SAX_HAR-LLM outperforms all symbolic baselines and most deep learning architectures, remaining competitive with state-of-the-art kernel-based approaches. Unlike competing methods that sacrifice transparency for accuracy, SAX_HAR-LLM’s predictions are fully auditable: an attention-based interpretability analysis reveals that the model’s predictive focus aligns with physically meaningful sensor axes and class-specific symbolic motifs, enabling decisions to be traced and validated in domain terms. These findings establish symbolic time-series representations as an interpretable and effective interface to LLMs — one where competitive performance and decision transparency are achieved simultaneously, at a modest but measurable cost in accuracy and inference overhead relative to purely numerical approaches.

Multivariate time series classification often requires balancing predictive performance with model simplicity and transparency, especially in domains constrained by interpretability or computational resources. This paper introduces a lightweight and interpretable symbolic framework for multivariate time series classification built upon the Symbolic Aggregate approXimation (SAX) representation. Despite SAX’s well-established merits in terms of dimensionality reduction, noise robustness, and interpretability, its practical effectiveness is highly sensitive to the choice of transformation parameters which are typically set heuristically. To address this limitation, we leverage the DIRECT global optimization algorithm to provide a principled, deterministic tuning framework that adapts these parameters to each dataset without resorting to ensemble or deep architectures. We propose four feature extraction strategies that vary in parameter sharing and channel aggregation, including a Multichannel Intelligent Icon (MII) frequency vector that captures cross-dimensional dependencies. Experiments on 24 benchmark datasets from the UEA archive demonstrate accuracy competitive to strong baselines with significant runtime advantages over computationally demanding alternatives, validated through Friedman and Nemenyi post-hoc statistical analysis. The proposed framework is particularly well-suited for applications where efficiency, trustworthiness, and explainability are essential, demonstrating that globally optimized symbolic representations constitute a viable alternative to more complex models for transparent multivariate time series classification.

The growing demand for eXplainable AI (XAI) has renewed interest in fuzzy cognitive maps (FCMs) due to their interpretability, causal transparency, and hybrid intelligence capabilities. However, current FCM explanation methods either overlook their dynamic behavior or limit themselves to feature attribution. Counterfactual explanations, which describe the minimal input changes required to alter outcomes, address this gap but remain largely unexplored in these models. The only existing FCM-specific approach relies on fuzzy discretization and predefined rules, producing causally invalid and overly conservative explanations. On the other hand, generic model-agnostic methods assume feature independence and suffer from instability, restrictive assumptions, and high computational costs. To overcome these limitations, this article presents the counterfactuals via the backpropagation (CF-BP) algorithm, a first backpropagation-based counterfactual explanation method for quasi-nonlinear FCMs (q-FCMs), which is a generalization of traditional FCMs that resolves convergence issues. CF-BP exploits the similarity between q-FCMs’ recurrent reasoning and neural network forward propagation, using exact analytical gradients to generate precise, causally consistent, and robust counterfactual explanations within the continuous state space of the model. Extensive evaluations, including hyperparameter sensitivity analysis and benchmarking against eight state-of-the-art (SOTA) model-agnostic methods, confirm the superior performance of the proposed method across key counterfactual quality metrics.

Metaverse brings a new era in educational settings, providing a wide range of services, such as VR-and XR-based skill training and AI-driven individualised learning experiences. While educational metaverses unlock unique opportunities for enriched learning experiences, they concurrently raise security and privacy concerns due to their heterogeneous nature. In this work, we provide a concise overview of the potentials of metaverse technologies within education and how their vulnerabilities could hinder them, formulating critical research questions encompassing a wide spectrum of concerns. In addition, we provide an overview of research works promising a trustworthy metaverse environment and propose potential remedies to address the identified challenges. By exploring these issues, this work aims to contribute to the development of trustworthy educational metaverse environments.

Leading a sedentary lifestyle is becoming a significant public health issue, but a way of dealing with this issue is Physical Activity (PA). Regular PA is proven to help prevent and manage chronic diseases, maintain healthy body weight, and improve fitness. However, starting any PA after a long sedentary lifestyle carries risks of injuries, and a smoother transition with appropriate fitness programs is required. The rapidly growing number of smartphone users has given birth to broad-spectrum apps that use built-in sensors and collect data to provide insights about health and fitness. This work presents a fitness mobile application developed following the World Health Organization (WHO) guidelines for the Physical Activity Readiness Questionnaire for Everyone (PAR-Q+) self-screening tool utilizing built-in mobile sensors.

This paper describes new interactive mental models applied to the pursuit of Construction Management (CM) project opportunities using Fuzzy Cognitive Maps (FCM). The CM-FCM models provide a basis for new decision support tools capable of providing Construction Industry Practitioners (CIP) support throughout the Project Life Cycle (PLC). It presents two novel CM-FCM models based on real-world construction engineering and management experience, specifically designed to support key decisions in the PLC. The interactive CM-FCM validates the application of FCM to this domain and demonstrate a method capable of helping manage the complexity and uncertainty inherent in construction management. The models offer a foundation for interactive intelligent decision support tools to assist with construction management.

The proposed hybrid algorithm aims at defining an FCM by the information carried by a large dataset about a specific problem, taking into consideration the BCO and FPO principles. The link between them is represented by the application of the DB-Scan clustering technique allowing to identify the right number of cluster without knowing it a priopri. The hybridisation highlights the efficacy of the algorithm in estimating the correlations among the factors involved for a specific problem, with low RMSE and computational time, demonstrated by the case study example.

In the last years, increased attention has been paid to the use of experts' opinions as concerned with the decision-making process and problem solving. Several approaches for aggregating them have been analysed but the literature review has highlighted that the experts' credibility in many cases is neglected. Thus, the present study aims at developing a novel approach for Fuzzy Cognitive Maps modelling, described in the literature as the unique methodology able to aggregate individuals knowledge and beliefs, combining a group decision-making technique, such as the Aggregation of Individual Evaluations one, and a structured credibility evaluation method according to the expertise areas of each expert. In particular, using a numerical example, the comparison between the traditional outcomes and those of the proposed method, highlights how the second one allows to obtain results more concordant with the single experts judgements than the first one, criticized of smoothing the experts opinion excessively.

This paper presents an extension of a hybrid method for modelling Fuzzy Cognitive Maps (FCMs), which combines human expert knowledge with existing recorded information and historical data. The proposed approach at first step extracts information from any type of dataset, extracts useful information and transforms it into knowledge in a form suitable to structure a FCM. The proposed approach is still dependent on experts' knowledge but it support them by providing the information and supportive data inferred from the dataset. Moreover, the approach highlights, through a supportive numerical example, that experts' knowledge and information from the dataset cannot be considered separately, since their combined use compensates possible errors in knowledge. In particular, it can be used to model a complex Decision Support Systems, and it also allows to identify the accuracy of the considered models.

This paper explores the possibility of creating an automatic method for assessing the condition of induction motor circuits fed by inverters. The stator current and magnetic flux are processed in the frequency domain and a feature selection stage is employed to pinpoint the most informative components to further be fed to a classifier that performs the assessment of the motor circuit. The results are promising, indicating that short circuit detection as well as quantification is feasible using noninvasive techniques.

The ever-increasing number of electronic content stored in digital libraries requires a significant amount of effort in cataloguing and has led to self-deposit solutions where the authors submit and publish their own digital records. Even in self-deposit, going through the abstract and assigning subject terms or keywords is a time consuming and expensive process, yet crucial for the metadata quality of the record that affects retrieval. Therefore, an automatic, or even a semi-automatic process that can recommend topics for a new entry is of huge practical value. A system that can address that has to rely basically on two components, one component for efficiently representing the relevant information of the new document and one component for recommending an appropriate set of topics based on the representation of the previous stage. In this work, different candidate solutions for both components are investigated and compared. For the first stage both distributed Document to Vector (doc2vec) and conventional Bag of Words (BoW) components are employed, while for the latter two different transformation approaches from the field of multi-label classification are compared. For the comparison, a collection of Ph.D. abstracts (~19000 documents) from the MIT Libraries Dspace repository is used suggesting that different combinations can provide high quality solutions.

The significant role of predicting weather conditions in daily life, the new era of innovative machine learning approaches along with the availability of high volumes of data and high computer performance capabilities, creates increasing perspectives for novel improved short-range forecasting of main meteorological parameters. Among the various algorithms for forecasting parameters, ensemble learning approaches are able to generate simple models which provide accurate predictions for regression problems. The advantage of ensembles with respect to single models is that they perform remarkably well for a variety of problems. The main aim of this ongoing research is to provide some preliminary assessment of the applicability of ensemble learning for wind speed forecasting. In this work, forecasting results of a single and two ensemble models are presented and compared.

Predicting the desirability and number of student applications to universities is a challenging and dynamic undertaking. Student applications are affected by a variety of factors ranging from those related to the university itself and its surrounding community, to factors up to the national level, including perceptions of international relations. This paper proposes a way of modeling this complex task and the related decision-making using fuzzy cognitive maps (FCM). The process includes determination of the key concepts and factors affecting the desirability of a university for a representative set of cohort groups. Cohort groups allow us to reuse model concepts and linguistic variables while customizing the relationships to create a custom, dynamic model based on a variety of target population subsets. This paper introduces a reusable process for developing models based on cohort groups, and presents three customized FCM models that can be used to explore a range of possibilities and provide a FCM model for prediction and planning concerning university desirability and selection.

In this paper, a data-driven approach for the classification of simultaneously occurring faults in an induction motor is presented. The problem is treated as a multilabel classification problem, with each label corresponding to one specific fault. The faulty conditions examined include the existence of a broken bar fault and the presence of mixed eccentricity with various degrees of static and dynamic eccentricity, while three “problem transformation” methods are tested and compared. For the feature extraction stage, the start-up current is exploited using two well-known time-frequency (scale) transformations. This is the first time that a multilabel framework is used for the diagnosis of co-occurring fault conditions using information coming from the start-up current of induction motors. The efficiency of the proposed approach is validated using simulation data with promising results irrespective of the selected time-frequency transformation.

In this article a data driven approach for the classification of simultaneously occurring faults in an induction motor is presented. The problem is treated as a multi-label classification problem with each label corresponding to one specific fault, using the power-set approach. The faulty conditions examined, include the existence of a broken bar fault and the presence of mixed eccentricity with various degrees of static and dynamic eccentricity. For the feature extraction stage, the time-frequency representation, resulting from the application of the short time Fourier transform of the start-up current is exploited. The proposed approach is validated using simulation data with promising results.

Fuzzy Cognitive Maps have been widely used for modeling complex systems but time and evolution of the system has not sufficiently been illustrated and taken into consideration within the FCM model. Time is a substantial aspect for any application because factors determining the behavior of the system evolve over time; they affect and change the route of any evolution of the system. This work further explains and justifies Timed Fuzzy Cognitive Maps (T-FCMs), an extension of the known soft computing technique of FCMs that can handle uncertainty to infer a result. T-FCM has been introduced to take into consideration the time evolution of any system and it provides intermediate modeling results. This work also introduces the combination of T-FCMs with Hidden Markov Model (HMM) to create an integrated system which always reach a decision, as HMM are called in the case that T-FCM do not converge to an acceptable state and HMM suggests the most probable state (decision-concept).

Taking time into consideration in the FCM model and the subsequent numerical calculation is a hot FCM issue. This work investigates an approach for introducing the time in Fuzzy Cognitive Maps (FCM) framework and it proposes the Timed-FCM (T-FCM) that incorporates time steps in their structure and constitutes an extension of FCMs. The proposed T-FCMs take into consideration the time evolution and the fact that the parameters of any system change with the time and it accepts intermediate states and results of the modeled system. In any problem, its factors change differently over time and so their influence as is presented on the interconnections between the concepts are time dependent.

This work presents a novel integrated approach on how to develop Fuzzy Cognitive Maps combining human expert knowledge with existing recorded information and historical data. The proposed approach aims to extract information from unstructured data and to transform it into knowledge in the form of a FCM. The proposed approach still is dependent on experts but providing them with more information and supportive data, in the form of particular evidence-based information available in the literature in order to better justify their selections. The proposed approach is inherit to the FCM and its structure the extracted knowledge and it can be used to model a complex system such as the Medical Decision Support Systems.

Implementation of unsupervised induction motor condition monitoring systems has drawn an increasing attention recently among motor drives manufacturers. In the case of soft-starters the possibility of incorporating fault detection features to their conventional functions provides an added value to those elements. Design and development of advanced algorithms that are able to automatically detect and alert about possible failures without requiring continuous human inspection is a challenging research goal. In this paper, an algorithm for the automatic detection of rotor damages in induction motors in the case of soft starting is proposed. The twofold approach relies, first, on the application of a time-frequency transform to the starting current signal and, second, on a pattern recognition stage based on the treatment of the time-frequency representation as a symbolic sequence. The innovation of this work is the implementation of the proposed approach for the automatic detection of rotor cage faults in soft-started motors. The experimental results prove the usefulness of the approach for the automatic detection of such faults and its potential for possible future implementation in soft-started machines.

One of the most common deficiencies of currently existing induction motor fault diagnosis techniques is their lack of automatization. Many of them rely on the qualitative interpretation of the results, a fact that requires significant user expertise, and that makes their implementation in portable condition monitoring devices difficult. In this paper, we present an automated method for the detection of the number of broken bars of an induction motor. The method is based on the transient analysis of the start-up current using wavelet approximation signal that isolates a characteristic component that emerges once a rotor bar is broken. After the isolation of this component, a number of stages are applied that transform the continuous-valued signal into a discrete one. Subsequently, an intelligent icon-like approach is applied for condensing the relative information into a representation that can be easily manipulated by a nearest neighbor classifier. The approach is tested using simulation as well as experimental data, achieving high-classification accuracy.

Eventual failures in induction machines may lead to catastrophic consequences in terms of economic costs for the companies. The development of reliable systems for fault detection that enable to diagnose a wide range of faults is a motivation of many researchers worldwide. In this context, non-invasive condition monitoring strategies have drawn special attention since they do not require interfering with the operation process of the machine. Though the analysis of the motor currents has proven to be a reliable, non-invasive methodology to detect some of the faults (especially when assessing the rotor condition), it lacks reliability for the diagnosis of other faults (e.g. bearing faults). The infrared thermography has proven to be an excellent, non-invasive tool that can complement the diagnosis reached with the motor current analysis, especially for some specific faults. However, there are still some pending issues regarding its application to induction motor faults diagnosis, such as the lack of automation or the extraction of reliable fault indicators based on the infrared data. This paper proposes a methodology that intends to provide a solution to the first issue: a method based on image segmentation is employed to detect several failures in an automated way. Four specific faults are analyzed: bearing faults, fan failures, rotor bar breakages and stator unbalance. The results show the potential of the technique to automatically identify the fault present in the machine.

Missing data cause serious problem for automatic evaluation of the fetal heart rate(FHR) series. In this work we present an algorithm to surpress this problem. More specifically, an adaptive approach is proposed based on two steps. The first step concerns the reconstruction step where we obtain an estimate of the missing data using an empirical dictionary. The second step consists from the construction of the dictionary using the updated values from the first step. The above two steps are applied iteratively until convergence. The method adapts each time the dictionary and the reconstructed time series to the new information that we gain. Results on real and simulated experiments have shown the usefullness of our approach. More specifically, a comparison with cubic spline interpolation is performed and have shown that the proposed approach achieved 4 to 9dB better reconstruction ability.

In this paper we propose the use of Intelligent Icons for both automatic assessment and representation of asynchronous machines' condition. The method focuses on the analysis of the start-up current for the isolation of a component that is able to pinpoint faulty signatures. The analysis is based on the application of Empirical Mode Decomposition (EMD) which acts as an adaptive filter during the start up and subsequently on the application of Symbolic Aggregate approXimation (SAX) for the transformation of the extracted component into a symbolic representation. Using this symbolic representation, an automated detection procedure can be developed that discriminates between faulty and normal conditions using an intelligent Icons approach while at the same time the information can be presented to the user in a more intuitive way.

Noisy optimization problems arise very often in real-life applications. A common practice to tackle problems characterized by uncertainties, is the re-evaluation of the objective function at every point of interest for a fixed number of replications. The obtained objective values are then averaged and their mean is considered as the approximation of the actual objective value. However, this approach can prove inefficient, allocating replications to unpromising candidate solutions. We propose a hybrid approach that integrates the established Particle Swarm Optimization algorithm with the Reinforcement Learning approach to efficiently tackle noisy problems by intelligently allocating the available computational budget. Two variants of the proposed approach, based on different selection schemes, are assessed and compared against the typical alternative of equal sampling. The results are reported and analyzed, offering significant evidence regarding the potential of the proposed approach.

This paper introduces a novel approach based on signal processing methods to extract features from speech signals and based on them to detect a specific type of articulation disorders. Articulation, in effect, is the specific and characteristic way that an individual produces the speech sounds. Empirical Mode Decomposition and the Hilbert Huang transform are applied to the speech signal in order to calculate the marginal spectrum of the signal. The marginal spectrum is subsequently subject to a mel-cepstrum like processing to extract features which are fed to a neural network classifier responsible for the identification of the articulation disorder. Our preliminary results suggest that this approach is very promising for the detection of the disorder under study.

Acoustic emission (AE) is one of the most important non-destructive testing (NDT) methods for materials and constructions. Using AE testing, the location of a single event (crack) can be classified efficiently into three typical areas in a ship hull. The problem is a typical classification problem based on the use of features extracted from piezo-sensors' signal. As in most classification problems, the extraction and selection of the most appropriate set of features plays a major role in the overall performance of the system. In this research work we investigate the use of an evolutionary algorithm to extract new features from a set of primitive features in a lower dimensional space through a linear transformation. These features are subsequently fed into a probabilistic neural network (PNN) that performs the classification. In simulation experiments, where a stiffened plate model (SPM) is partially sank into water, the localization rate in noisy environments outperforms a work, where a feature selection phase alone was used before the classification phase. The proposed hybrid computational intelligent approach shows the potential merit of using it in real life situations where the signal is distorted by noise.

This paper presents a new hybrid modeling methodology for the complex decision making processes. It extends previous work on competitive fuzzy cognitive maps for medical decision support systems by complementing them with genetic algorithms methods. The synergy of these methodologies is accomplished by a new proposed algorithm that leads to more dependable advanced medical decision support systems that are suitable to handle situations where the decisions are not clearly distinct. The methodology developed here is applied successfully to model and test a differential diagnosis problem from the speech pathology area for the diagnosis of language impairments.

Intrapartum electronic fetal monitoring (EFM) is an indispensable means for fetal surveillance. However, the early enthusiasm was followed by scepticism, since the introduction of EFM in every day practise resulted in an increase in operative deliveries. Nevertheless the drawbacks of EFM relate not so much to the technique itself but more to the difficulties in reading and interpreting the fetal heart rate (FHR). In an attempt to develop more objective means to analyse the FHR recordings and compensate for the different levels of expertise among clinicians, computerized systems have been developed during the last 2 decades. In this work, we present an approach to automatic classification of FHR tracings belonging to hypoxic and normal newborns. The classification is performed using a set of parameters extracted from the FHR signal and two hidden Markov models (one for each class). The results are satisfactory indicating that the FHR convey much more information than what is conventionally used.

Fuzzy cognitive maps is a hybrid method based on fuzzy systems and neural networks and belonging in soft computing. The methodology of developing fuzzy cognitive maps (FCMs) is easily adaptable and relies on human expert experience and knowledge, but it exhibits weaknesses in utilization of learning methods. The external intervention (typically from experts) for the determination of FCM parameters and the convergence to undesired steady states are significant FCM deficiencies. Thus, it is necessary to overcome these deficiencies in order to improve efficiency and robustness of FCM. Weight adaptation methods can alleviate these problems by allowing the creation of less error prone FCMs where causal links-weights are adjusted through a learning process.

This research deals with the soft computing methodology of fuzzy cognitive map (FCM). Here a mathematical description of FCM is presented and a new methodology based on fuzzy logic techniques for developing the FCM is examined. The capability and usefulness of FCM in modeling complex systems and the application of FCM to modeling and describing the behavior of a heat exchanger system is presented. The applicability of FCM to model the supervisor of complex systems is discussed and the FCM-supervisor for evaluating the performance of a system is constructed; simulation results are presented and discussed.