A Stable and Adaptable Machine Learning Framework for Phishing Detection

Adeno-associated viral (AAV) vectors for gene therapy are becoming integral to modern medicine, providing therapeutic options for diseases once deemed incurable. Currently, optimizing viral vector purification is a critical bottleneck in the gene therapy industry, impacting product efficacy and safety as well as accessibility and cost to patients. Traditional optimization methods are resource-intensive and often fail to adjust the purification process parameters to maximize the resulting product yield and quality. To address this challenge, we developed a machine learning framework that leverages Bayesian optimization to systematically refine affinity chromatography parameters (sample load, flow rate, and the formulation of chromatographic media) to improve AAV purification. The efficiency of this closed-loop workflow in iteratively optimizing the vector’s yield, purity, and transduction efficiency was demonstrated by purifying clinically-relevant serotypes AAV2, AAV5, and AAV9 from HEK293 cell lysates using the affinity adsorbent AAVidity. We show that three cycles of Bayesian optimization elevated yields from a baseline of 70% to 99%, while reducing host-cell impurities by 230-to-400-fold across all serotypes. The optimized parameters consistently produced vectors with high purity and preserved high transduction activity, essential for therapeutic efficacy and safety, demonstrating serotype versatility – a key challenge in AAV manufacturing. By streamlining parameter optimization and enhancing productivity, our adaptive machine learning framework accelerates process development and reduces costs, advancing the accessibility and clinical translation of AAV-based gene therapies.

Adeno-associated viral (AAV) vectors for gene therapy are becoming integral to modern medicine, providing therapeutic options for diseases once deemed incurable. Currently, viral vector purification is a critical bottleneck in the gene therapy industry, impacting product efficacy and safety as well as accessibility and cost to patients. Traditional methods for improving viral vector purity are resource-intensive and often fail to adjust the purification process parameters to maximize the resulting product yield and quality. To address this challenge, we developed a machine learning framework that leverages Bayesian optimization to systematically refine affinity chromatography parameters (sample load, flow rate, and the formulation of chromatographic media) to improve AAV purification. The efficiency of this closed-loop workflow in iteratively optimizing the vector's yield, purity, and transduction efficiency was demonstrated by purifying clinically relevant serotypes AAV2, AAV5, AAV6, and AAV9 from HEK293 cell lysates using the affinity adsorbent AvXcel. We show that in three (or fewer) cycles of Bayesian optimization, we elevated yields from a baseline of 70% to a remarkable 97%-99%, while reducing host cell impurities by 230- to 400-fold across all serotypes. Performing the purification process with optimized parameters consistently produced vectors with high purity and preserved high transduction activity, essential for therapeutic efficacy and safety, demonstrating the applicability of the framework across multiple serotypes-a key challenge in AAV manufacturing. This study represents the first reported application of closed-loop, data-driven Bayesian optimization for enhancing AAV productivity and quality at the affinity capture step, with demonstrated transferability of historical purification data and process knowledge. The proposed adaptive machine learning framework is efficient and applicable across serotypes, enabling rapid process development, reduced costs, and advancing the accessibility and clinical translation of AAV-based gene therapies.

From a system’s theoretical point of view, adaptive learning systems (ALS) for education and training contain in their core – in a simplified form – closed feedback control loops in which the control is determined by the measured users’ performance. Improving this performance can increase the learning outcome, especially for critical disciplines such as education or training for disaster risk management. However, for this special form of intelligent (e-learning) assistance systems, learning theories and behavioral models have to be considered, e.g., game flow theory, cognition models, or learning models. The research question is how adaptive interactive learning environments (ILE) such as serious games and computer simulations can be characterized and analyzed to determine optimal adaptation strategies. Adaptive learning environments should adapt to the context-related needs of the user in order to ensure and optimize learning success, especially for disaster management training. This contribution presents a concept for an interoperable, adaptive ILE framework which follows control theory and its models, contributing to the state of the art for adaptive games or simulations in disaster risk management.

Occupant behavior prediction has become a significant part of smart building technologies. This research emphasizes on HVAC (Heating Ventilation and Air Conditioning) system as its operation is crucial in terms of both energy usage and user comfort. In previous researches, comfort prediction is developed using the filled questionnaires from occupants. The prediction output is employed to control the mechanical components of the system. On/Off control is mostly developed using rule-based distance triggering that turns on/off the system when occupants enter/exit a specific perimeter. This article proposes a plug-and-play HVAC control system with an adaptive cloud machine learning (ML) framework that utilizes room current ambient conditions together with the historical adjustment log to automatically adjust room temperature according to the user comfort using Support Vector Classification (SVC). This system can also turn on air conditioner automatically using predictive control which implements GPS location and the occupant's data on the Artificial Neural Network (ANN). Furthermore, cloud deployment can solve problems in processing power and storage of an end device, also provide scalability for future development. Findings show that an adaptive algorithm in the ML framework can perform well, even when the occupant's behavior changes slowly.

Deep learning has shown remarkable potential for complex signal recognition, yet its deployment in noisy real-world environments remains a major challenge. This study presents an adaptive deep learning framework for acoustic signal detection, demonstrated through the identification of cetacean whistles in the Strait of Gibraltar, a mandatory migratory bottleneck where several species coexist with intense maritime traffic and complex acoustic conditions. Passive acoustic monitoring (PAM) enables long-term assessment of marine mammal presence and behavior, yet its application remains challenging due to overlapping anthropogenic and environmental sounds and the large volumes of data generated. The proposed framework combines transfer learning, semi-supervised iterative fine-tuning, and confidence-threshold calibration to automate the analysis of PAM data and improve robustness in acoustically variable conditions. Pre-trained bioacoustic models (BirdNET and Perch) were repurposed as feature extractors and coupled with custom classifiers to assess generalization across domains. Model performance was evaluated using standard deep learning metrics, including accuracy, recall, and F1-score. Models were validated using a clean benchmark dataset and a real-world deployment dataset characterized by substantial anthropogenic and geophonic noise. While baseline models achieved over 0.95 accuracy on clean data, their performance degraded under realistic noise conditions (whistle F1-score ≤ 0.10). The best-performing model, fine-tuned with local data and semi-supervised iterative validation, achieved an F1-score of 0.88 and improved recall across deployments. Confidence-threshold optimization further enhanced adaptability, supporting both automated and expert-assisted monitoring workflows. The final model was tested across three independent acoustic deployments spanning different seasons and locations. Results demonstrated robust whistle detection with minimal annotation effort. This work provides a reproducible, AI-driven framework for signal detection in noisy environments, demonstrating how adaptive learning and threshold calibration can extend the applicability of deep neural networks to real-world acoustic monitoring and contribute to the development of automated, ecologically meaningful soundscape analysis systems. • A semi-supervised deep learning framework is presented for acoustic monitoring. • Integrated BirdNET and Perch embeddings with iterative expert validation. • Adaptive threshold optimization (0.79–0.84) balances automation with validation. • Model-assisted annotation reduces labeling effort while preserving quality over 1319 h. • Scalability of an AI-based approach is assessed in a high-impact PAM context.

In the oil and gas drilling industry, optimizing key operational parameters—such as weight on bit (WOB), rate of penetration (ROP), and Revolutions Per Minute (RPM)— is essential for maximizing drilling efficiency, minimizing costs, and ensuring operational safety. Drilling environments are highly dynamic, influenced by constantly changing geological formations, mud properties, and downhole conditions. This research addresses the critical need for real-time optimization of drilling parameters. Despite significant advancements in drilling automation, many existing systems still lack the adaptability to respond effectively to changing downhole conditions and transparency in decision-making. This study proposes an adaptive machine learning (ML) and optimization model framework to bridge this gap. The goal is to both validate the performance of models in dynamic drilling environments and assess how interpretability methods can foster operator trust. Using the dataset provided, the framework simulates real-time adaptability through an offline, condition-segmented analysis approach. The dataset is divided into subsets representing distinct drilling conditions, such as variations in depth or Mud Weight (MW). For each condition, regression and time series models are trained to predict the rate of penetration, and an optimization algorithm is then implemented to continuously provide optimum controllable parameters, for maximum penetration rates within safe Formation Pressure ranges. The results show significant improvements in drilling efficiency and cost reduction, demonstrating the potential of the proposed framework for real-time optimization. Additionally, the study highlights the importance of integrating ML and optimization techniques in drilling operations, enabling more accurate and informed decision-making, and paving the way for further advancements in autonomous drilling systems and digital transformation in the oil and gas industry, with potential applications in various drilling scenarios.

Dyslexia is a learning disorder in which individuals have significant reading difficulties. Previous studies found that using machine learning techniques in content supplements is vital in adapting the course concepts to the learners' educational level. However, to the best of our knowledge, no research objectively applied machine learning methods to adaptive content generation. This study introduces an adaptive reinforcement learning framework known as RALF through Cellular Learning Automata (CLA) to generate content automatically for students with dyslexia. At first, RALF generates online alphabet models as a simplified font. CLA structure learns each rule of character generation through the reinforcement learning cycle asynchronously. Second, Persian words are generated algorithmically. This process also considers each character's state to decide the alphabet cursiveness and the cells' response to the environment. Finally, RALF can generate long texts and sentences using the embedded word-formation algorithm. The spaces between words are proceeds through the CLA neighboring states. Besides, RALF provides word pronunciation and several exams and games to improve the learning performance of people with dyslexia. The proposed reinforcement learning tool enhances students' learning rate with dyslexia by almost 27% compared to the face-to-face approach. The findings of this research show the applicability of this approach in dyslexia treatment during Lockdown of COVID-19.

An adaptive machine learning framework integrating large language models to assess and enhance emotional intelligence in adolescents.

Learning style detection is essential for developing adaptive learning tailored to learner preferences. Extensive research has been conducted to address the limitations of direct learner engagement, behavioral-based approaches, and the lack of learner-teacher interaction caused by the shift to online learning. Many learning style models have been widely used and evaluated for their performance in supporting adaptive learning. However, implementing Multiple Intelligence Learning Styles (MILS) remains limited in online learning due to low detection performance. This study employs machine learning approaches to enhance the detection performance of Multiple Intelligence Learning Styles. This study proposes an integrated machine learning framework to enhance MILS classification by combining data preprocessing, log-modulus transformation, and class imbalance handling via Synthetic Minority Over-Sampling Technique (SMOTE), Adaptive Synthetic Sampling Approach (ADASYN), and SMOTE-Tomek. Recursive Feature Elimination (RFE) with Support Vector Machine (SVM) is used for feature selection while eight machine learning models are used for classification. Interactions between learners and the Learning Management System (LMS) comprise the dataset, which are labeled using the Multiple Intelligence Inventory. The SVM model using ADASYN outperforms previous studies and other models, achieving the highest F1-Score of 89%, according to experimental data using 5-fold cross-validation. Statistical tests (Shapiro-Wilk, Analysis of variance (ANOVA), and paired t-tests) confirm significant differences in performance between models. The proposed approach demonstrates enhanced detection performance and can be adapted to broader e-learning applications, supporting adaptive and personalized learning systems.

Adaptive and Explainable Machine Learning Framework for Real-Time Credit Scoring and Financial Fraud Detection with Privacy-Preserving Intelligence

This chapter presents a novel domain correction and adaptive extreme learning machines framework (DC-AELM) with transferring capability to solve the drift and interference problem of E-nose. The framework consists of two parts: (1) domain correction (DC) that makes the distributions of two domains close; (2) adaptive extreme learning machine (AELM) that learns a transferable classifier at decision level. This method is motivated by the idea of transfer learning, especially from the perspective of domain correction and decision making, to realize the knowledge transfer for interference suppression and drift compensation. Experiments on a background interference dataset and a public benchmark sensor drift dataset via E-nose verify the effectiveness of the proposed DC-AELM method.

Dynamic source domain selection: An adaptive EEG transfer learning framework for mitigating negative transfer.

Objective. In brain-computer interfaces (BCIs) that utilize motor imagery (MI), minimizing calibration time has become increasingly critical for real-world applications. Recently, transfer learning (TL) has been shown to effectively reduce the calibration time in MI-BCIs. However, variations in data distribution among subjects can significantly influence the performance of TL in MI-BCIs. Approach. We propose a cross-dataset adaptive domain selection transfer learning framework that integrates domain selection, data alignment, and an enhanced common spatial pattern (CSP) algorithm. Our approach uses a huge dataset of 109 subjects as the source domain. We begin by identifying non-BCI illiterate subjects from this huge dataset, then determine the source domain subjects most closely aligned with the target subjects using maximum mean discrepancy. After undergoing Euclidean alignment processing, features are extracted by multiple composite CSP. The final classification is carried out using the support vector machine. Main results. Our findings indicate that the proposed technique outperforms existing methods, achieving classification accuracies of 75.05% and 76.82% in two cross-dataset experiments, respectively. Significance. By reducing the need for extensive training data, yet maintaining high accuracy, our method optimizes the practical implementation of MI-BCIs.

A hyperspectral imaging system enhanced by machine learning is developed to monitor impurities in liquid dielectrics of power equipment in real time. This system integrates advanced hyperspectral imaging with machine learning techniques to address the critical challenge of detecting impurities in transformer oils and coolants. Research focuses on developing a field-deployable solution for identifying metallic particles (Fe, Cu) and carbon-based contaminants that degrade dielectric performance in smart grids. The system architecture combines a custom-designed hyperspectral imaging module (390–1010 nm range, 2.5 nm resolution) with a hierarchical ensemble architecture integrating a 1D CNN (5-layer), SVM with RBF kernel, XGBoost, LightGBM, and a GRU meta-learner, optimized via Bayesian tuning and enhanced by preprocessing steps like first-derivative analysis and continuum removal. The experimental results demonstrate robust detection performance across different dielectric liquids, with the system achieving 87.79% accuracy for coolant impurity identification and an higher 91.17% accuracy for transformer oil contaminant detection. The technology establishes a comprehensive spectral fingerprint database specifically for smart grid liquid dielectrics, enabling precise material characterization through unique spectral signatures. The study makes significant contributions to smart grid maintenance by: developing the first real-time, non-invasive monitoring solution for liquid dielectrics; creating an adaptive machine learning framework optimized for spectral analysis; and demonstrating practical implementation in smart grid environments. The system’s modular design and edge-computing compatibility facilitate broader applications in energy infrastructure monitoring.

This study proposes AIEAM, an AI-Enhanced Adaptive Learning Framework designed to support personalized education in higher education. The framework integrates cognitive, metacognitive, and ethical components into a modular architecture comprising five core modules: learner profiling, adaptive path generation, real-time feedback and assessment, resource recommendation, and an ethical and trust layer. Grounded in established learning theories, AIEAM addresses the limitations of existing systems that often lack pedagogical transparency, multidimensional adaptivity, and ethical oversight. To validate the framework, two real-world systems—Squirrel AI and Carnegie Learning (MATHia)—are analyzed against AIEAM’s design principles. The analysis reveals strong alignment in functional components while highlighting gaps in metacognitive support and ethical integration. AIEAM contributes both a conceptual model and practical roadmap for developing adaptive learning environments that are scalable, interpretable, and pedagogically sound. The framework provides a foundation for future research and implementation strategies in AI-supported higher education learning systems.