In aquatic monitoring, predicting water quality conditions helps describe how environmental variables may change and supports management-oriented analysis. Machine learning methods are increasingly used to learn nonlinear relationships from observed water quality data. Beyond short-term prediction, multi-horizon forecasting has attracted growing interest to anticipate system evolution over extended lead times. However, existing studies remain fragmented in terms of forecasting strategies, model design, and evaluation practices, making it difficult to assess robustness across horizons. This paper reviews machine learning-based multi-horizon prediction approaches in water quality forecasting as a sustainability-oriented context. Rather than focusing on individual models, the review analyzes how multi-horizon prediction problems are formulated, how forecasting strategies and model choices influence horizon-dependent behavior, and how performance is evaluated across lead times. To support clearer comparison and reproducibility, we introduce a structured taxonomy of standard forecasting strategies and propose a minimal, actionable evaluation protocol for consistent multi-horizon assessment. The analysis highlights recurring challenges, including error accumulation with increasing horizons, sensitivity to data quality, rising model complexity, limited generalization across sites, and inconsistent evaluation protocols. The paper concludes by outlining key directions for future research toward more robust, interpretable, and scalable multi-horizon prediction frameworks for sustainable environmental monitoring.

Fish production forecasting is essential for sustainable fisheries management, particularly in data-scarce regions. This study evaluates the performance of machine learning models (Random Forest Regression, XGBoost), time series models (SARIMA, ETS), and hybrid approaches to predict monthly fish production in Sfax, Tunisia, using hyperlocal climate data from 2000 to 2023. While traditional models achieved moderate accuracy, a hybrid model combining Exponential Smoothing (ETS) and XGBoost proved the most effective. By integrating time series dynamics with climate-informed features, this model achieved strong performance, with an $R^{2}$ of 0.97 on the training set, and $R^{2}=0.83$ on the test set, and a low MAE of 0.01. The findings demonstrate the hybrid model's ability to capture both seasonal patterns and complex non-linear relationships, providing a reliable forecasting tool to support sustainable fisheries management in the face of climatic variability.

Marine pollution is a major global concern, with eutrophication emerging as a particularly alarming consequence that severely threatens coastal ecosystems worldwide. This study presents a framework to enhance the prediction of marine eutrophication in the Bizerte Lagoon, Tunisia, an area heavily impacted by anthropogenic activities. Our primary objective is to predict Chlorophyll-a (Chl-a) concentration, a key bio-indicator for the Trophic Index (TRIX). We employed and compared several machine learning models using physical-chemical data from 2012, such as regression techniques inlcuding Random Forest Regression (RFR), Support Vector Regression (SVR), and XGBoost, and time-series models, specifically the Autoregressive Integrated Moving Average (ARIMA) and Seasonal ARIMA (SARIMA) models. The results demonstrated that time-series models that account for temporal dynamics performed significantly better than static regression models. The SARIMA model yielded the best performance, achieving an $\mathbf{R}^{2}$ of 0.63, an MSE of 1.09, and an MAE of 0.82. The predicted Chl-a values were then used to calculate the TRIX index, which indicated a medium level of eutrophication in the lagoon. These findings highlight the effectiveness of time-series modeling for improving water quality assessments and provide a valuable tool for environmental agencies to mitigate the impacts of marine pollution.

The early detection and precise classification of Karenia selliformis, a toxic microalga responsible for harmful algal blooms (HABs), are crucial for the protection of marine ecosystems and the management of aquatic resources, particularly in Tunisia. In this study, we propose an artificial intelligence (AI)-based framework that combines YOLOv8 for detection with convolutional neural networks (CNNs) for classification. We systematically evaluated four CNN architectures-VGG16, InceptionV3, EfficientNetB0, and ResNet-50-to identify the most effective model to integrate with YOLOv8. Among the models tested, ResNet50 demonstrated superior performance, achieving a precision of 93 %, making it the most suitable choice for the accurate classification of Karenia selliformis. These findings underline the value of deep learning and object detection techniques for improving the real-time monitoring of harmful microalgae and highlight their potential for use in early warning systems for marine ecosystem protection efforts in Tunisia.

This paper explores the integration of machine learning with the trophic index (TRIX) to assess eutrophication and water quality, with a specific focus on the Mediterranean region and other ecologically sensitive areas. While various indices, such as the Water Quality Index and Pollution Index, exist, they often fall short of the TRIX index’s flexibility and specificity in aquatic eutrophication assessment. This paper highlights the benefits of combining machine learning models with TRIX for a more predictive, data-driven approach to water quality management.

Belief change is an important topic of knowledge representation and reasoning in artificial intelligence. Within the logical framework, the AGM approach has become a standard and various belief change operations have been considered. While revision, contraction and updating have given rise to a great deal of work, erasure has so far attracted less interest. Erasure is to contraction what update is to revision.This article deals with the study of erasure within the framework of propositional logic. It extends Katsuno and Mendelzon’s approach with additional postulates capturing the minimality of change and proposes two representation theorems for erasure operators, one in terms of total preorders on interpretations, the other in terms of partial preorders on interpretations. Finally, it completes the work of Caridroit, Konieczny and Marquis for contraction by proposing a new representation theorem for contraction operators in terms of partial preorders on interpretations.

Industry 4.0 combines virtual and physical worlds, enabling cyber-physical systems (CPS) through computer vision. The ability of computers to understand and interpret visual data opens up numerous possibilities for innovation and efficiency in various domains. Face shape classification is crucial for selecting eyelashes, hairstyles, makeup, and glasses frames based on guidelines and personal preferences. Automated face shape identification systems can alleviate this time and effort. Existing methods face challenges due to face geometry and variations. Computer vision has been affected by the appearance of profound learning due to the need to characterize or translate the substance of regular pictures, the availability of large datasets, and the creation of convolution layers that could be run quickly on GPUs with remarkable accuracy. This work presents an approach for classifying face shapes into five types: oval, heart, oblong, square, and round.

Incorporating the latest advancements in computer vision and virtual reality (VR) technology, this paper explores the fusion of computer vision and augmented reality, resulting in groundbreaking innovations applicable across various industries. This study introduces an advanced virtual eyewear try-on system that transcends its traditional application in the fashion sector. This cutting-edge system harnesses the power of precise facial and ocular recognition algorithms, sophisticated image manipulation techniques, geometric transformations, and real-time visual rendering. As a result, users are immersed in a lifelike environment, enabling them to virtually experiment with a diverse range of eyewear frames. Beyond its notable impact on the online retail experience, this technology opens up exciting possibilities in the realm of healthcare.

As the world grapples with the ongoing COVID-19 pandemic, machine learning models have emerged as indispensable tools for predicting and analyzing positive cases. Through their ability to process vast amounts of data and identify complex patterns, these models offer valuable insights to guide public health strategies. However, challenges such as data quality, ethical considerations, and integration into existing frameworks remain. As researchers continue to refine and optimize machine learning models, their potential to aid in the fight against COVID-19 becomes increasingly promising. By leveraging the power of artificial intelligence, we can better anticipate the course of the pandemic, allocate resources efficiently, and ultimately save lives. This work is a simple Comparing different machine learning techniques (Random Forest RF, LSTM, ARIMA, Support Vector Machine SVM and extreme Gradient Boosting XGBoost) for predicting the number of COVID-19 positive cases, fatalities, and individuals who have received the vaccine. The best-performing model that gives motivating results can be used for the prediction of any epidemic disease. This study can help researchers analyze and predict COVID-19 in Maghrebian Countries.

This paper deals with belief base revision, a form of belief change which consists in restoring consistency with the intention of incorporating a new piece of information from the environment, while minimally modifying the agent's belief state represented by a finite set of propositional formulas. In an effort to guarantee more reliability and rationality for real applications while performing revision, we come up with the idea of credible belief base revision. We define two new formula-based revision operators using tools offered by evidence theory. These operators, uniformly presented in the same spirit as [92], [13], stem from consistent sub-bases maximal with respect to credibility instead of set inclusion or cardinality. Logical properties and productivity of inference from these operators are established, along with complexity results.

Recently, belief change within the framework of fragments of propositional logic has gained attention. In the context of revision, it has been proposed to refine existing operators so that they operate within propositional fragments and that the result of revision remains in the fragment under consideration. Later, this notion of refinement was generalized to belief change operators. Whereas refinement allowed one to define concrete rational operators adapted to propositional fragments in the context of revision and update, it has to be specified for contraction and erasure. We propose a specific notion of refinement for contraction and erasure operators, called reasonable refinement. This allows us to provide refined contraction and erasure operators that satisfy the basic postulates. We study the logical properties of reasonable refinement of two model-based contraction operators and two model-based erasure operators. Our approach is not limited to the Horn fragment but applicable to many fragments of propositional logic, like Krom and affine fragments.

This paper deals with belief base revision that is a form of belief change consisting of the incorporation of new facts into an agent's beliefs represented by a finite set of propositional formulas. In the aim to guarantee more reliability and rationality for real applications while performing revision, we propose the idea of credible belief base revision yielding to define two new formula-based revision operators using the suitable tools offered by evidence theory. These operators, uniformly presented in the same spirit of others in [9], stem from consistent subbases maximal with respect to credibility instead of set inclusion and cardinality. Moreover, in between these two extremes operators, evidence theory let us shed some light on a compromise operator avoiding losing initial beliefs to the maximum extent possible. Its idea captures maximal consistent sets stemming from all possible intersections of maximal consistent subbases. An illustration of all these operators and a comparison with others are inverstigated by examples.

The present paper describes briefly a project idea in progress about the evolvement of individuals' opinions, beliefs and perceptions on social networks (such as Facebook, Twitter, Instagram, youtube...) which is a thorny subject that has whetted nowadays the curiosity of a hulk of researchers from various disciplines. For this purpose, differently from a lot of works in the literature, we rely on logical knowledge representation tools in order to investigate the belief merging operation of Artificial Intelligence (AI). The major objective of this project is to provide efficient operator for merging heterogeneous, inconsistent and uncertain multiple sources information in the context of social networks taking into account the fact that opinion can be formed and developed through the concept of social influence with its two forms (informational social influence and normative social influence) and the concept of social trust. We intend thus through this research work presenting an adaptative version to our context of an approach [7] expressed thanks to Answer Set Programming (ASP) paradigm with stable model semantics. It is worth to say that our approach profits from the impressive volume data produced by users in social networks about a particular topic by learning from opinions, beliefs and perceptions that their freinds/neighbors share and therefore allows to use this kind of data to extract initial opinions, and to validate the proposed opinions merging process allowing even the prediction of users' behaviors.

Recently, Creignou et al. (Theory Comput. Syst. 2017), introduced the class DelayFPT into parameterised complexity theory in order to capture the notion of efficiently solvable parameterised enumeration problems. In this paper, we propose a framework for parameterised ordered enumeration and will show how to obtain enumeration algorithms running with an FPT delay in the context of general modification problems. We study these problems considering two different orders of solutions, namely, lexicographic order and order by size. Furthermore, we present two generic algorithmic strategies. The first one is based on the well-known principle of self-reducibility and is used in the context of lexicographic order. The second one shows that the existence of a neighbourhood structure among the solutions implies the existence of an algorithm running with FPT delay which outputs all solutions ordered non-decreasingly by their size.

This paper deals with the complexity of model checking for belief base revision. We extend the study initiated by Liberatore & Schaerf and introduce two new belief base revision operators stemming from consistent subbases maximal with respect to cardinality. We establish the complexity of the model checking problem for various operators within the framework of propositional logic as well as in the Horn fragment.

Recently, belief change within the framework of fragments of propositional logic has gained attention. In the context of revision it has been proposed to refine existing operators so that they operate within propositional fragments, and that the result of revision remains in the fragment under consideration. In this paper we generalize this notion of refinement to belief change operators. Whereas the notion of refinement allowed one to define concrete rational operators adapted to propositional fragments in the context of revision and update, it has to be specified for contraction. We propose a specific notion of refinement for contraction operators, called reasonable refinement. This allows us to provide refined contraction operators that satisfy the basic postulates for contraction. We study the logical properties of reasonable refinements of two well-known model-based contraction operators. Our approach is not limited to the Horn fragment but applicable to many fragments of propositional logic, like Horn, Krom and affine fragments.

Recently, belief change within the framework of fragments of propositional logic has gained attention. Previous works focused on belief revision, belief merging, and on belief contraction in the Horn fragment. The problem of belief update within the framework of fragments of propositional logic has been neglected so far. In the same spirit as a previous extension of belief revision to propositional fragments, we propose a general approach to define new update operators derived from existing ones such that the result of update remains in the fragment under consideration. Our approach is not limited to the Horn fragment but applicable to many fragments of propositional logic, like Horn, Krom and affine fragments. We study the logical properties of the proposed operators in terms of the KM’s postulates satisfaction and highlight differences between revision and update in this context.

Recently, belief change within the framework of fragments of propositional logic has gained attention. Previous works focused on belief contraction and belief revision mainly on the Horn fragment. However, the problem of belief update within the framework of fragments of propositional logic has been neglected so far. This paper presents a general approach to define new update operators derived from existing ones such that the result of update remains in the fragment under consideration. Our approach is not limited to the case of the Horn fragment but applicable to any fragment of propositional logic characterized by a closure property of the sets of models of their formulas. We study the logical properties of the proposed operators in terms of K. and M.’s postulates satisfaction.

The classes Delay-FPT and Total-FPT recently have been introduced into parameterized complexity in order to capture the notion of efficiently solvable parameterized enumeration problems. In this paper we focus on ordered enumeration and will show how to obtain Delay-FPT and Total-FPT enumeration algorithms for several important problems. We propose a generic algorithmic strategy, combining well-known principles stemming from both parameterized algorithmics and enumeration, which shows that, under certain preconditions, the existence of a so-called neighbourhood function among the solutions implies the existence of a Delay-FPT algorithm which outputs all ordered solutions. In many cases, the cornerstone to obtain such a neighbourhood function is a Total-FPT algorithm that outputs all minimal solutions. This strategy is formalized in the context of graph modification problems, and shown to be applicable to numerous other kinds of problems.