Accurate forecasting of time series data is essential in many fields. However, real-world time series are often characterized by noise, non-stationarity and multiscale temporal dependencies, which collectively reduce forecasting performance. To address these challenges, MultiScaleWave, a deep learning framework based on time series decomposition, is proposed for univariate forecasting. The MultiScaleWave model first applies multi-level discrete wavelet transforms to decompose the series into multiscale temporal components. Each component is modeled by a granularity-adaptive module, and the outputs are then fused to generate an informative representation for final forecasting. The MultiScaleWave model has been validated on benchmark datasets and achieves superior performance compared to competitive baselines. The results demonstrate the effectiveness and generalizability of the proposed approach.

Asthma is a chronic respiratory condition shaped by environmental, physiological, and behavioral factors. Accurate prediction of asthma severity is vital for personalized care and reducing exacerbations. While Machine Learning (ML) has been widely explored in asthma prediction, many existing models lack generalizability, robustness, or comprehensive integration of multiple predictors, limiting their clinical applicability. This study presents a robust ML-based model for classifying asthma severity by incorporating diverse patient and environmental features. A supervised learning approach was employed using a publicly available dataset of 1,010 records with 14 features, including demographics, clinical symptoms, and environmental indicators (temperature, wind speed, and humidity). The dataset was pre-processed and stratified to balance severity classes. Three models—Decision Tree, Support Vector Machine (SVM), and Random Forest (RF) were evaluated using standard metrics: accuracy, precision, recall, F1-score, and ROC AUC. Among them, the RF model showed superior performance, achieving 96.70% accuracy, a 0.9668 F1-score, and a 0.9831 ROC AUC. Feature importance analysis highlighted environmental factors, particularly temperature and humidity, as key predictors of asthma severity. These results underscore RF's effectiveness in providing accurate, interpretable predictions and addressing limitations of earlier models. The proposed model offers a data-driven framework for real-time severity forecasting, supporting early interventions and personalized treatment. It holds promise for integration into clinical decision support systems, thereby enhancing asthma management and optimizing healthcare resource use. It is therefore recommended that the proposed model be operationalized within clinical triage protocols, embedded into electronic health record (EHR) infrastructures, or integrated into mobile health (mHealth) applications to facilitate data-driven, proactive asthma management across diverse care settings in Nigeria.

FedROS-RandNN: A lightweight online Federated Learning framework for glycemic forecasting using Randomized Neural Networks

A generalized approach for predicting and mitigating total dissolved gas supersaturation at data-scarce dams.

A physics-guided sensor-to-model framework for real-time estimation and near-future forecasting of soil moisture

A hybrid reinforcement learning framework for adaptive multi-horizon electricity load forecasting: The DWRNet approach

Accurate streamflow prediction plays a vital role in water management and flood mitigation. However, conventional deep learning models often fail to simultaneously capture short-term variability and long-term dependencies, particularly in univariate time series common in hydrological settings. To address these challenges, we propose the Multi-Resolution Adaptive Channel Fusion Transformer Encoder LSTM (MR-ACF-TE-LSTM)-hybrid architecture designed to improve predictive accuracy and interpretability by modeling temporal patterns at multiple scales. The model constructs pseudo-multivariate inputs from lagged observations, statistical summaries, and seasonal indicators, which are dynamically fused through an adaptive attention-based mechanism. Before making a prediction with LSTM, this fused representation are encoded by a transformer and then refined over time. Extensive experiments on three benchmark streamflow datasets demonstrate that MR-ACF-TE-LSTM consistently outperforms baseline models, including Transformer, Transformer-LSTM, and Bayesian CNN, achieving lower RMSE and higher R² scores. Ablation and comparative analyses demonstrate that MR-ACF-TE-LSTM attains the minimal RMSE across all FMSs, with multi-resolution yielding enhancements of up to 13% and overall RMSE reductions varying from 28% to 48% in comparison to baseline and state-of-the-art models. In comparison to baseline models, the RMSE at Dereevi FMS improved by 39%, from 1.436 to 0.874. Cross-dataset evaluations further highlight the model's robustness and generalization capabilities across heterogeneous catchments. Ablation studies confirm the critical contributions of multi-resolution inputs and adaptive fusion, while attention weight visualizations reveal the model's ability to selectively focus on temporally inputs. These findings establish MR-ACF-TE-LSTM as a powerful and interpretable framework for univariate hydrological forecasting.

Background: Markets of cocoon silk suffer the price instability and farmers have high levels of economic uncertainty. The traditional forecasting methods do not reflect the seasonality, the environment and market driven interactions that affect price changes. The unpredictability of demand due to fluctuation of demand, climatic conditions and variation in production cycles makes the task of income planning extremely burdensome since the production of cocoons sustains millions of small farmers. The paper will look into the possibility of a more data-driven machine-learning model to capture these complicated market dynamics. Methods: This study presents a leakage-free machine learning framework for early forecasting of cocoon silk prices using a Random Forest regression model. To maintain realistic forecasting conditions, contemporaneous price variables were excluded and predictions were generated solely through lagged historical modal prices together with pre-available environmental and management indicators. The dataset was properly preprocessed and temporally ordered and model evaluation was done using a chronological train-test split. Performance was assessed using error-based metrics and explained variance, showing the non-stationary nature of agricultural price series. Result: The revised model achieved a Mean Absolute Error (MAE) of 2965.14, a Root Mean Square Error (RMSE) of 5985.60 and an explained variance (R2) of 0.0903. While the explained variance is modest, this behavior is appropriate for early forecasting in volatile agricultural markets when data leakage is eliminated. Feature contribution analysis reveals that short-term historical prices have the strongest influence, with environmental and management aspects providing secondary but meaningful effects. A simulation-based assessment indicates a potential 12-18% revenue improvement, resulting from informed selling decisions enabled by short-horizon price forecasts. This gain shows lower exposure to post-harvest price troughs and improved market selection rather than precise peak-price prediction.

The increasing complexity and variability of electricity consumption in modern power systems have highlighted the need for intelligent, real-time forecasting solutions to ensure reliable, cost-efficient, and sustainable grid oper ations. This research presents Smart Cast, an AI-driven load forecasting frame work that integrates Long Short-Term Memory (LSTM) networks with Gradient Boosting models to capture both nonlinear temporal patterns and external influ encing factors such as weather dynamics. The system leverages historical smart meter data along with real-time sensor inputs to generate accurate short- and mid term demand predictions. In addition, Smart Cast incorporates anomaly detection techniques to proactively identify irregular consumption behaviors that may in dicate faults, energy theft, or operational inefficiencies. Outputs are delivered through an interactive web dashboard, enabling utility operators and industrial consumers to visualize demand behavior and receive cost-aware recommenda tions for optimized energy usage. By enhancing grid planning, reducing opera tional risks, and increasing renewable integration potential, Smart Cast presents a scalable and practical solution to advancing smart-grid intelligence and overall energy sustainability.

The optimized Grok algorithm can significantly improve the accuracy of time series analysis and understanding the dynamics of climate change. Fine-tuned Grok architecture can be used to monitor and analyze climate processes. The main aim is to analyze the Fine-tuned Grok architecture for research on climate change, world ecology, carbon dioxide growth, and carbon funds. The global challenges of climate change and ecological degradation demand innovative analytical approaches capable of processing vast, multivariate, and non-linear datasets. Concurrently, the global financial system, deeply intertwined with energy transitions and sustainable development, requires sophisticated tools for risk assessment and investment strategy in a changing world. Fine-tuned Grok architecture model helps to plan strategies for adaptation to climate change by calculating the optimal allocation of resources, taking into account risks and reducing losses. Due to its ability to respond quickly to new conditions, the system will be able to quickly adjust evacuation plans, deploy protective structures, and distribute assistance to affected regions. The use of artificial intelligence significantly expands the capabilities of the scientific community and authorities in monitoring, assessing, and managing climate change. The optimized Fine-tuned Grok architecture opens the way to a new level of informed decision-making about climate change and ensuring the safety of our future generations.

Demand forecasting plays a critical role in supply chain management, enabling suppliers, manufacturers, and retailers to synchronise operations and enhance overall efficiency. Despite extensive research on time series forecast model selection, choosing the most appropriate forecasting model for a given time series remains a complex challenge, particularly in volatile and uncertain environments. The increasing availability of data and the emergence of new forecasting methods have introduced greater complexity, making automated model selection essential for improving forecasting accuracy and decision-making in supply chain operations. This study proposes an automated demand forecast model selection framework that integrates a broad range of statistical and machine learning models. A key feature of the framework is the optimisation of hyperparameters across all models, ensuring each method is fine-tuned for optimal performance. The approach is validated on the M3 monthly dataset, where it outperforms all previously submitted methods, demonstrating significant improvements in forecast accuracy. Additionally, the methodology is tested in a real-world supply chain setting, further showcasing its effectiveness in handling complex and dynamic demand patterns. By enhancing forecast accuracy and reducing the reliance on manual model selection, this research provides an efficient decision support system for supply chain demand forecasting in fast-changing supply chain environments.

Urban flood risk in dam-regulated catchments is a dynamic and escalating challenge, driven by the interplay of hydrological modifications, land use change, and climate variability. This study develops and validates a forecasting framework that integrates Artificial Neural Networks (ANN) with geospatial analysis to project future urban flood risk in the Shiroro Dam catchment, Niger State, Nigeria. The framework synergistically combines urban growth simulation, climate change Scenarios, and an ANN model trained on topographic, climatic, land cover, and soil moisture variables. Analysis of multi-temporal Landsat imagery (2014-2024) revealed significant landscape transformation, including substantial agricultural loss and water body expansion due to dam impoundment. The ANN model demonstrated superior predictive performance (accuracy: 91.6%; Kappa: 82%) compared to traditional GIS-overlay methods. It identified 35.06% of the study area as highly vulnerable to flooding, with population densities in risk zones projected to reach 7,318 persons/km² by 2034. The study provides a robust, transferable tool for proactive flood risk management, emphasizing the need for integrated land-use planning and offering a scalable methodology for other dam-affected catchments.

Sleep is important to human health. To improve sleep quality, recorded EEG signals have been utilized for automated sleep staging either in real time during sleep or after sleep. However, because previous approaches classify events that have already occurred rather than forecasting the future, their effectiveness is limited for personalized sleep management. This study proposes the sleep stage forecaster with sleep EEG tokenizer (SSF-SET) framework, which predicts the future sleep-stage sequence using only earlier EEG of the current sleep. SET combines a multi-branch transformer for epoch-level representations, an LSTM-based sequence encoder-decoder, and quantization to convert continuous EEG features into sleep-informative tokens. This quantization makes an information bottleneck that reveals the latent transition structure and suppresses artifacts, enabling reliable next-stage prediction. The decoder-only transformer SSF is first pretrained for next-token prediction with causal attention, then fine-tuned via reinforcement learning that uses sequence-level macro-F1 and token-consistency rewards; throughout it does not access future EEG at inference. With subject-wise cross-validation on the SleepEDF20 and SleepEDF78 datasets, SSF-SET consistently outperformed direct forecasting and forecasting with predicted sleep stages. On SleepEDF20, accuracy was 0.596 and macro-F1 was 0.516. In addition, we achieved an accuracy of 0.611 and a macro-F1 of 0.537 on SleepEDF78. These results show that quantized sleep EEG tokens are effective for autoregressive prediction and demonstrate that future sleep stages can be predicted without future EEG. We believe that SSF-SET is an important component for developing closed-loop and personalized sleep interventions that can act before disruptive transitions occur, and we expect it to improve sleep quality.

Introduction Short-range forecasting of tropical cyclone intensity remains challenging because storm evolution is governed by complex, nonlinear dynamics. Accurate 24-h wind speed prediction is particularly important for operational decision-making but is difficult to achieve using traditional approaches alone. Sequence-based deep learning models offer a data-driven way to learn temporal dependencies in intensity evolution from historical observations. This study proposes a transparent and computationally efficient sequence-to-one deep learning framework as a proof of concept for short-term cyclone intensity forecasting. Methods This study presents a sequence-based deep learning framework for 24-h wind speed forecasting, formulated as a sequence-to-one prediction task using historical cyclone observations. Multivariate best-track data from the International Best Track Archive for Climate Stewardship (IBTrACS) are organized into fixed-length temporal sequences and modeled using a Long Short-Term Memory (LSTM) network to capture temporal dependencies in intensity evolution. Results Model performance is evaluated using normalized error metrics, demonstrating low relative prediction error and stable learning of short-term intensity trends. Predicted wind speeds are further mapped to standard cyclone intensity categories to enable qualitative assessment of categorical consistency. Case-based analyses and interpretability experiments suggest that recent intensity history and physically meaningful track-related variables play a dominant role in the model’s forecasts, consistent with established understanding of tropical cyclone behavior. Discussion The proposed framework emphasizes transparency and computational efficiency and is intended as a proof-of-concept demonstration of sequence-based modeling for cyclone intensity forecasting. While the current implementation does not incorporate environmental predictors or provide direct quantitative comparison with baseline forecasting methods, it establishes a foundation for future extensions involving additional predictors, broader basin coverage, and systematic benchmarking against operational approaches.

This paper describes the Quarterly Macro Forecasting Framework (QMFF), which is a novel approach to macroeconomic policy analysis and forecasting. At the core of this framework is a Quarterly Projection Model, with main behavioral equations quantified as deviations of real variables from their trends. The model comprises a simultaneous system of calibrated equations that cover the main sectors of the Financial Programming and Policies framework, as well as key accounting restrictions within and across sectors. By explicitly accounting for trends observed in real variables and relative prices, the framework ensures a balanced growth path and constant expenditure shares relative to nominal GDP. The QMFF is sufficiently flexible to be adapted to different monetary frameworks and exchange rate regimes. After describing the framework, the paper illustrates its implementation for policy analysis and forecasting on the case of Bosnia and Herzegovina. The empirical work with the model includes not only calibration but also testing the model’s dynamic and in-sample simulation properties.

An Integrated VMD-CNN-GRU Framework for Multi-Horizon Forecasting of Water Quality Dynamics

A parallel attention-based framework for multi-step multivariate runoff forecasting in mountainous watersheds: Wuyuan case study

The main objective of this study are to develop an intelligent forecasting for Electric Vehicle Charging Station (EVCS) and to significantly enhance the accuracy of energy consumption forecasting in renewable integrated smart grid environments. These objectives are achieved through solving the following tasks: implementing data preprocessing to handle missing values, remove outliers and eliminate inconsistent observations for improving dataset reliability; performing feature engineering for generating meaningful temporal and derived variables that strengthen model interpretability; and carrying out detailed Exploratory Data Analysis (EDA) for extracting statistical trends, recognize correlations and uncover hidden temporal dependencies in energy consuption behaviour. Structure on the preparatory stages, a hybrid Deep Learning (DL) approach using a Radial Basis Spiking Net (ARBSNet) is developed by combining radial basis kernal (RBF) with temporal behavior of Spiking Neural Networks (SNN), enhanced with attention mechanisms for capturing non-linear fluctuations and time varying required pattern. The most important results obtained from Python based experiments highlight enhancement in forcasting performance, with the proposed model achieving a Mean Squared Error (MSE) of 0.1183, a Mean Absolute Error (MAE) of 0.2694, a Root Mean Squared Error (RMSE) of 0.3439, and an overall prediction accuracy reaching a 2 R score of 0.99. The significant of the results lies in their ability to support predictive energy allocation, optimize load balancing strategies and improve grid stability. By providing highly dependable demand forecasts for charging infrastructure, the proposed framework contributes to the sustainable integration of electric mobility within future smart energy systems.

Wind energy has gained significant attention as a clean, renewable resource due to fossil fuels' environmental impact. Accurate wind speed forecasting is essential to address variability and intermittency challenges. Current forecasting difficulties arise from wind speed's high susceptibility to meteorological conditions. This study proposes a GA-based ensemble framework that combines forecasting models using genetic algorithms. We systematically compared 14 models: linear models (AR, ARMA), advanced neural networks (MLP, RBF), hybrid models, and ensembles. Models were evaluated using minute-by-minute data from five major Brazilian cities: Brasília, Florianópolis, Petrolina, Natal, and São Luís. Key findings include: I) Superior Performance: The proposed framework achieved MSE values from 0.0802 to 0.9020 and MAE values from 0.1970 to 0.6140 across all datasets; II) Robust Prediction: values ranged from 0.7139 to 0.8723, demonstrating strong predictive capability; III) Statistical Validation: Friedman test ( ) confirmed significant differences with perfect rank stability across all locations; IV) High Scalability: Runtimes ranged from 58,077.3 to 77,815.7 s, determined by the base model combination; and V) Computational Efficiency: One-step-ahead forecasting requires only 0.0003 s for weighting and combination.

Abstract Accurate forecasting of precious metal prices is increasingly critical in modern financial markets, as these metals function as industrial commodities and as strategic financial instruments for portfolio diversification and risk management. Although recent advances in financial technology have produced a range of forecasting approaches from traditional econometric methods to sophisticated deep learning models—the complex dynamics of metal prices continue to challenge existing methodologies. This paper introduces a significant innovation in financial forecasting by revealing and leveraging previously unrecognized pattern relationships in decomposed time series data. Our comprehensive analysis of metal price dynamics reveals distinct grouped patterns in decomposed time series components, challenging the conventional assumption of independence in current forecasting methods. Based on these insights, we propose the pattern-guided forecasting framework (PGFF), which enhances forecasting accuracy by leveraging cross-dimensional pattern relationships in decomposed time series. Our framework employs a novel two-stage approach: first, categorizing decomposed time series based on their temporal characteristics and autocorrelation patterns; then, implementing cross-dimensional forecasting to capture complex market dynamics. Empirical analysis of four major precious metals demonstrates that PGFF consistently outperforms existing forecasting frameworks, offering significant implications for investment decision-making and portfolio management in modern financial markets.