Time series forecasting, particularly in the domain of stock prices, is a significant challenge but benefits from the availability of openly accessible data. Our work focuses mainly on (although not limited to) univariate time series forecasting of monthly or daily stock prices, predicting one step ahead. We developed an innovative pipeline that combines fuzzification with Automated Machine Learning, achieving improved forecasting performance. Unlike previous literature, we revise the binomial Fuzzy Time Series and machine learning algorithm , including a classification task (formally motivating it), and involving unused features of fuzzy sets . Thanks to the type of aggregation of the fuzzified data, the approach has the potential to preserve interpretability, unlike most machine learning based approaches . Using several financial datasets, in addition to preliminary experiments on chaotic time series, we found evidence of significantly improved performance in most cases. This study contributes to further understanding of the intersection between fuzzy logic and Automated Machine Learning, particularly in the context of time series forecasting, offering a promising direction for future research.

In an era where image manipulation tools are widely accessible, detecting digital image forgery has become increasingly challenging. Image forgery detection is a critical concern in cybersecurity, particularly within IoT-based applications and digital forensics. This paper proposes an automated approach for image forgery detection, integrating Fuzzy C-Means clustering with TPOT, an automated machine learning framework, to optimize the detection pipeline. The approach is evaluated on four publicly available datasets: FIDAC, FID, CoMoFoD, and CASIA_V2. Experimental results demonstrate that the proposed approach outperforms conventional CNN and VGG models, achieving 96.83% accuracy on CASIA_V2 and 95.33% on CoMoFoD. These findings highlight the potential of AutoML-based solutions to enhance forgery detection.

Abstract As neural network-based localization algorithms are becoming popular, there is a need to shorten the training time and the localization time for sustainability and efficiency purposes. To address such issues, the fuzzy transform (or F-transform for short) is employed here for the first time in a neural network-based localization algorithm. The F-transform is a dimensionality reduction method, which has found several applications over the last decade, but it has not been well explored in the form of a prepending layer to a neural network. In this respect, some properties (including the computational cost) of the F-transformed neural scheme are formally discussed here. The performance of the neural network-based approach with and without F-transform, and with a state-of-the-art reduction technique, i.e. the principal component analysis, is evaluated first on simulated data and then on publicly available real-world data. Different neural network architectures have been tried jointly with the above-mentioned reduction techniques. The numerical experiments show the excellent performance of the proposed fuzzy transform-based approach, which can ensure considerable savings in training time and query response time, without significant losses in accuracy.

In this study, we present a fuzzy contrast-based model that classifies mental patient authored text into different symptoms by using an attention network for position-weighted words. Then, the mental data are labeled using the trained embedding. After that, the lexicons of the attention network are extended to allow the use of transfer learning methods. Our proposed approach classifies weighted attention words using similarity as well as contrast sets. The fuzzy model then classifies mental health data into different groups. To illustrate the performance of the proposed model, the approach is compared with the non-embedding as well as standard approaches. From the demonstrated results, the feature vector has a high Receiver Operating Characteristic Curve (ROC)-curve of 0.82 for 9 different symptom problems.

In the literature, a popular dissimilarity measure between each pair of fuzzy data is defined as a weighted sum of the squared Euclidean distances among the centers and the spreads. The flexible weights introduced by D’Urso and Giordani in [Computational Statistics & Data Analysis 50 (2006) 1496–1523] may be helpful in certain situations as shown in that paper. The flexible weights are obtained by the minimization algorithm, and a larger weight is given to the center distance than the spread one.In the present study, the weights are proposed to be fixed and equal to 0.5, which can contribute to higher performances of clustering methods in some cases. For instance, where the distance among the spreads is longer than that among the centers, i.e., the centers do not play a relevant role while the spreads play the predominant role, it is clear that a clustering method that takes the advantage of equal weights would outperform the other. Indeed, when there is no external condition, choosing the weights fixed and equal to 0.5 would be the best strategy.To deeply investigate our claim, we provide a wide simulation study to show whether and under what circumstances the flexible weights are (not) helpful. The results furnished by numerical experiments on both simulated and benchmark datasets demonstrate the merit of using the fixed and equal to 0.5 weights suggested in the present paper.

In this article, we introduce a variant of the adaptive network-based fuzzy inference system (ANFIS). The proposed variant does not use backpropagation and grid partitioning, but the least-squares method with fractional Tikhonov regularization. The fractional regularization is a generalization of the standard regularization and is applied here to the learning process of the ANFIS scheme for the first time. This results in a simpler rule base, with a low number of rules, allowing to handle problems with many input variables with relatively low computational time while keeping high accuracy. We present new theoretical results on the fractional Tikhonov regularization. Such results are the basis for a formal discussion on how much the choice of a different architecture, resulting in a different matrix in the least-squares minimization, could affect the accuracy. We perform several numerical experiments on benchmark examples, first to assess the impact of the fractional regularization on the accuracy and then to compare our results against the most recent ones reported in the literature by other ANFIS-like or neuro-fuzzy systems. The numerical results show the good performance of the proposed approach.

In this paper, we introduce a variant of the Adaptive Network-based Fuzzy Inference System (ANFIS). The proposed variant does not use backpropagation and grid partitioning. Scatter partitioning is employed by complementing the least-squares method with Tikhonov regularization, both in standard and fractional version. The application example is the prediction of the burnt area in forest fires. We used two publicly available datasets for the numerical experiments. The results encourage further investigations.

In this article, analogies and differences between a type of fuzzy transform and a type of autoencoder, both based on a least-squares optimization, will be discussed. Such schemes have been recently introduced in the literature in different contexts. In particular, in this article, the data compression application will be considered. As it will be shown, the least-squares fuzzy transform can be regarded as a kind of autoencoder with a lower computational cost, without losing accuracy. The numerical comparison against existing results for the considered application shows the good performance of the fuzzy transform based approach.

In this paper, a fuzzy-based paradigm for data compression aimed at reducing the computational burden of data analysis in smart grids (SGs) is proposed. In the SG context, it is challenging achieving an efficient use of the channel communication bandwidth and a reduced need of the storage space for operational data. Thus, we discuss a fuzzy-based mathematical kernel which transforms the data into a new domain where their cardinality can be sensibly reduced, consequently allowing the development of more efficient data analysis algorithms. Detailed numerical results obtained from several test systems are presented and discussed in order to demonstrate the effectiveness of the proposed approach to handle SG operation problems.

Events that deal with Group Decision Making are continuously studied in order to provide a suitable representation of different opinions, with the aim of reaching the consensus of all experts involved in decision processes. In this paper, the authors, focusing on employees' evaluations inside Italian companies, propose an extension of a fuzzy consensus model dealing with a feedback process to guide the decisions. Precisely, a fuzzy logic approach is used to compute the importance degree of the experts considering, besides their experiences and roles, the profile of the resource to evaluate, i.e. a factor that indicates the working trend of the employee. This allows more fair evaluations of resources, as the importance of each expert also considers the behavior of employees during their whole working period. A case study, that focus on the evaluations inside a real Italian company, is useful to analyze the proposed approach.

In this paper, we present our results related to the definition of a methodology that combines augmented reality (AR) with semantic techniques for the creation of digital stories associated with museum exhibitions. In contrast to traditional AR approaches, we augment real-world elements by supplementing contents of a museum exhibition with additional inputs that provide new and different meanings. In this way we augment a cultural resource with respect to both its presentation and meaning. The methodology is framed in the cultural re-mediation theory and is grounded on a set of ontologies aimed at modelling a cultural resource and correlating it with external multimedia objects and resources. To provide an easy tool for the creation of museum narratives, the methodology makes use of a set of recognised practices widely adopted by museum curators that have been formalised through inference rules. The defined methodology has been experimented in a scenario related to Flemish paintings to validate the augmentation of cultural objects with two different approaches, the first basing on similarities and the second on dissimilarities.

This paper discusses a new computational scheme based on functional networks and applies it to the problem of classification and quantification of gas species in a mixture. A generalized functional network as a new classifier is proposed to improve the potentialities of the standard functional network classifier. Both methodology and learning algorithm are derived. The performance of this new classifier is examined by using experimental applications. A comparative study with the most common classification algorithms is carried out by showing the high-quality performance of the proposed classifier. The classifier interacts with some quantifiers, again based on functional networks and finite differences. The scheme of the quantifiers was previously proposed for single gas exposure applications and is here extended to the multigas case. Numerical results show that our approach behaves quite satisfactorily.