Adil El Makrani

Accurate stock market forecasting remains challenging due to the nonlinear, noisy, and volatile nature of financial markets. Beyond historical prices, market movements are influenced by financial news and investor sentiment. This study proposes a hybrid financial bidirectional encoder representations from transformers with convolutional neural network and long short-term memory (FinBERT-CNN-LSTM) model for forecasting the Standard & Poor’s 500 (S&P 500) index by integrating financial news sentiment with historical market data. FinBERT generates daily aggregated sentiment scores, which are combined with market features and processed through a CNN-LSTM architecture. The CNN captures local patterns and short-term interactions, while the LSTM models temporal dependencies in financial time series. Experimental results show that the proposed model outperforms benchmark models, achieving a root mean square error (RMSE) of 0.0119, mean absolute error (MAE) of 0.0091, mean absolute percentage error (MAPE) of 2.10%, and directional accuracy (DA) of 63.4%. It also achieves a cumulative return of 15.8% and a Sharpe ratio of 1.35. Rolling-window validation and Diebold–Mariano (DM) testing further support the statistical significance of the improvements. These findings demonstrate the effectiveness of combining financial sentiment and deep learning (DL) for stock market forecasting.

Supply chain digitalization has increased the need for reliable systems that support transparency, traceability, trust, and structured interorganizational information exchange. Electronic Data Interchange (EDI) remains widely used for standardized business transactions, while blockchain offers decentralized verification, data immutability, and stronger data governance. However, blockchain–EDI integration depends not only on technical compatibility but also on organizational capacity, partner alignment, financial resources, and regulatory preparedness. Existing studies mainly examine blockchain benefits and adoption barriers, with limited attention to readiness assessment before implementation. This study develops an analytical framework for evaluating organizational preparedness for blockchain–EDI integration in supply chains. Five readiness dimensions are identified from the literature: technological, organizational, partner, financial, and regulatory readiness. These dimensions are measured using a 0–5 scoring system, combined into a weighted readiness score, and linked to a logistic function that estimates integration success under different complexity levels. Deterministic simulation, Monte Carlo simulation, and sensitivity analysis are used to examine the model. The results show a nonlinear readiness-success relationship under assumed parameter values: low readiness is associated with limited estimated success, medium readiness forms a transition zone, and high readiness supports more stable estimated outcomes. The framework is positioned as a methodological readiness assessment model rather than an empirically validated predictive system. It provides a basis for future empirical calibration, pilot testing, and validation using organizational implementation data.

Facial expression recognition (FER) is essential in various domains such as healthcare, road safety, and marketing, where real-time emotional feedback is crucial. Despite advancements in controlled settings such as well-lit, frontal, and unobstructed conditions, FER still faces significant challenges in natural, unconstrained environments. One of the most difficult issues is the presence of occlusions, which obscure key facial features. To overcome this, multiple strategies have been proposed, generally falling into two categories: those focused on analyzing visible facial regions and those aimed at reconstructing hidden facial features. In this study, we present a variational autoencoder (VAE)-based solution designed to reconstruct facial features obscured by occlusions. Experimental results show our VAE model optimized with the structural similarity index measure (SSIM) cost function achieves superior performance, with recognition rates of 91.2% for eye occlusions and 89.7% for mouth occlusions. The SSIM-optimized VAE effectively reconstructs occlude facial features while preserving structural details, demonstrating significant improvements over conventional approaches. This VAE-based solution proves particularly robust for real-world scenarios involving common facial obstructions like masks or sunglasses, making it valuable for applications in healthcare monitoring, driver safety systems, and human-computer interaction.

Faced with the intensification of water shortages caused by climate change and human pressures, this chapter presents a hybrid architecture combining predictive modelling and extreme scenario generation to strengthen territorial resilience. The approach is based on the integration of time series transformers and deep generative models to anticipate the evolution of hydrological variables and simulate critical events. In the absence of sufficiently rich real data, a set of realistic synthetic data was constructed from hydrological distributions, seasonal cycles and stochastic processes, then qualitatively validated. This game allows a rigorous evaluation of predictive and generative performance. Results demonstrate the complementarity of the two dimensions to improve risk visualization and activation of early responses. The chapter also discusses technical limitations, ethical issues, and proposes concrete perspectives for integration into digital twins, warning systems or decision support tools for climate adaptation.