Machine Learning Methodology Research Articles

Optimizing Cybersecurity Incident Response via Adaptive Reinforcement Learning

Cybersecurity threats have evolved dramatically over the past few decades, requiring organizations to continuously improve their security posture. Traditional cybersecurity incident response (CIR) frameworks, which rely on predefined rules and heuristics, have shown significant limitations in addressing sophisticated and rapidly evolving cyberattacks. The increasing complexity of threat landscapes necessitates adaptive security mechanisms capable of learning and evolving in real time. This paper explores the potential of Adaptive Reinforcement Learning (ARL) as a mechanism to enhance cybersecurity incident response strategies. Reinforcement learning (RL), a subset of machine learning, is well-suited for dynamic decision-making scenarios, where optimal strategies emerge through iterative learning. By integrating adaptive RL techniques into CIR, cybersecurity professionals can develop response strategies that continuously refine themselves based on observed threats, attack vectors, and system vulnerabilities. The study first examines conventional CIR approaches, discussing their constraints in modern cybersecurity environments. A comprehensive literature review explores the existing machine learning methodologies applied to cybersecurity and the emerging role of reinforcement learning in security applications. The methodology section presents the design and implementation of an ARL-driven incident response framework, detailing the algorithmic foundation, data sources, and training methodology. The effectiveness of the proposed approach is validated through extensive simulations across different cyberattack scenarios. Results highlight the superior performance of adaptive RL models in minimizing response time, improving threat mitigation rates, and reducing false positives when compared to traditional rule-based and supervised learning approaches. In addition to analyzing the results, the paper discusses practical challenges in deploying RL-based cybersecurity frameworks, including computational overhead, adversarial learning risks, and the need for high-quality training data. Future research directions are explored, emphasizing the importance of integrating federated learning techniques, adversarial resilience mechanisms, and multi-agent reinforcement learning systems to further enhance cybersecurity defenses. This study contributes to the growing field of AI-driven cybersecurity by demonstrating how adaptive reinforcement learning can optimize decision-making processes in real-time incident response, ultimately paving the way for more intelligent and resilient cyber defense strategies.

Optimizing heart disease diagnosis with advanced machine learning models: a comparison of predictive performance

Cardiovascular disease is the leading cause of mortality globally, necessitating precise and prompt predictive instruments to enhance patient outcomes. In recent years, machine learning methodologies have demonstrated significant potential in enhancing the precision and efficacy of health-related predictions, especially in the identification of heart disease. The dataset used in this study came from the UC Irvine Machine Learning Repository and included data from Cleveland, Switzerland, Hungary, Long Beach, and Statlog. We selected seven of the 1,190 cases, each with 12 attributes, for analysis. We used different machine learning models, like Random Forest, K-Nearest Neighbors, Logistic Regression, Naïve Bayes, Gradient Boosting, AdaBoost, XGBoost, and Bagged Trees, to check performance using accuracy, precision, recall, F1-score, and ROC-AUC. K-fold cross-validation (K = 10, K = 5) was conducted to guarantee the robustness and generalizability of these models. Random Forest exhibited remarkable stability, attaining 94% accuracy with K = 10 and 92% with K = 5, whereas XGBoost had a minor decrease during cross-validation (90% for K = 10, 89% for K = 5). KNN demonstrated possible overfitting, evidenced by a notable decline in accuracy (71% for K = 10, 72% for K = 5). XGBoost and Bagged Trees achieved the highest accuracy of 93%, followed by Random Forest and KNN at 91%. Furthermore, Random Forest and Bagged Trees exhibited the highest ROC-AUC values at 95%, and XGBoost demonstrated a ROC-AUC of 94%. The results demonstrate the effectiveness of ensemble methods in predicting cardiac diseases, along with the potential for future advancement through the incorporation of hybrid models and advanced survival analysis techniques.

Student dropout prediction through machine learning optimization: insights from moodle log data

Student attrition and academic failure remain pervasive challenges in education, often occurring at substantial rates and posing considerable difficulties for timely identification and intervention. Learning management systems such as Moodle generate extensive datasets reflecting student interactions and enrollment patterns, presenting opportunities for predictive analytics. This study seeks to advance the field of dropout and failure prediction through the application of artificial intelligence with machine learning methodologies. In particular, we employed the CatBoost algorithm, trained on student activity logs from the Moodle platform. To mitigate the challenges posed by a limited and imbalanced dataset, we employed sophisticated data balancing techniques, such as Adaptive Synthetic Sampling, and conducted multi-objective hyperparameter optimization using the Non-dominated Sorting Genetic Algorithm II. We compared models trained on weekly log data against a single model trained on all weeks’ data. The proposed model trained with all weeks’ data demonstrated superior performance, showing significant improvements in F1-scores and recall, particularly for the minority class of at-risk students. For example, the model got an average F1-score across multiple weeks of approximately 0.8 in the holdout test. These findings underscore the potential of targeted machine learning approaches to facilitate early identification of at-risk students, thereby enabling timely interventions and improving educational outcomes.

Prioritizing sub-watersheds for soil and water resource conservation using multi-analytical approaches: a study of Shilabati River Basin, W.B., India

ABSTRACT Prioritizing watersheds is critical for effective river basin management, as it guides resource allocation, decision-making, and monitoring to enable long-term comprehensive natural resource management. The present research proposes multi-analytical approaches for prioritizing 23 sub-watersheds in the Shilabati River Basin (SRB) of West Bengal, India, with a focus on soil and water conservation and management. Using cutting-edge methods like remote sensing and geographic information systems, this study provides a thorough analysis of the basin's geomorphological, geohydrological, geoenvironmental characteristics like morphometric analysis (MA), principal component analysis (PCA), hypsometric analysis (HA), ARC-SWPT analysis, and land use/land cover (LULC) Analysis. Quantitative measurements of MA, including 12 morphometric characteristics were used to rank and prioritize sub-watersheds. PCA was employed to rank sub-watersheds based on highly correlated morphometric parameters. HA was utilized to assess erosion and deposition stages, providing insights into landform evolution. LULC analysis depicted land use patterns and their impact on soil and water conservation. Traditional watershed research techniques and machine learning methodologies were combined to manage and prioritize sub-watersheds for a range of socioeconomic and environmental goals. Common sub-watersheds were ultimately identified into three priority levels: low, moderate, and high, along with raking. High-priority sub-watersheds can be used by planners to organize land use management and conservation strategies to prevent flooding, and soil erosion while simultaneously conserving land.

Short-term salinity prediction for coastal areas of the Vietnamese Mekong Delta using various machine learning algorithms: a case study in Soc Trang Province

Saltwater intrusion has significant and diverse impacts on agriculture, freshwater resources, and the well-being of coastal communities. To effectively address this issue, precise models for predicting saltwater intrusion must be developed, as well as timely information for reaction planning. In this study, a spectrum of machine learning (ML) methodologies, specifically Random Forest Regression (RFR), Support Vector Regression (SVR), Long Short-Term Memory (LSTM), Artificial Neural Network (ANN), Extreme Gradient Boosting (XGBoost), and Ridge Regression (RR), was systematically employed to predict salinity levels within the coastal environs of the Mekong Delta, Vietnam. The input dataset comprised hourly salinity measurements from Tran De, Long Phu, Dai Ngai, and Soc Trang stations and hourly water-level data from Tran De station and hourly discharge data from the Can Tho hydrological station. The dataset was partitioned into two distinct sets for the purpose of model development and evaluation, employing a division ratio of 75% for training (constituting 8469 observations) and 25% for testing (comprising 2822 observations). The results indicate that ML models are suitable for short-term salinity prediction, with a forecasting time of up to 16 h in this area. These research findings highlight the potential of machine learning in addressing saltwater intrusion and provide valuable insights for developing appropriate response policies. By leveraging the strengths of these models and considering the optimal forecasting time, policymakers can make informed decisions and implement effective measures to mitigate the impacts of saltwater intrusion in the Mekong Delta.

Revealing gut microbiota biomarkers associated with melanoma immunotherapy response and key bacteria-fungi interaction relationships: evidence from metagenomics, machine learning, and SHAP methodology

IntroductionThe gut microbiota is associated with the response to immunotherapy in cutaneous melanoma (CM). However, gut fungal biomarkers and bacterial-fungal interactions have yet to be determined.MethodsMetagenomic sequencing data of stool samples collected before immunotherapy from three independent groups of European ancestry CM patients were collected. After characterizing the relative abundances of bacteria and fungi, Linear Discriminant Analysis Effect Size (LEfSe) analysis, Random Forest (RF) model construction, and SHapley Additive exPlanations (SHAP) methodology were applied to identify biomarkers and key bacterial-fungal interactions associated with immunotherapy responders in CM.ResultsDiversity analysis revealed significant differences in the bacterial and fungal composition between CM immunotherapy responders and non-responders. LEfSe analysis identified 45 bacterial and 4 fungal taxa as potential biomarkers. After constructing the RF model, the AUC of models built using bacterial and fungal data separately were 0.64 and 0.65, respectively. However, when bacterial and fungal data were combined, the AUC of the merged model increased to 0.71. In the merged model, the following taxa were identified as important biomarkers: Romboutsia, Endomicrobium, Aggregatilinea, Candidatus Moduliflexus, Colwellia, Akkermansia, Mucispirillum, and Rutstroemia, which were associated with responders, whereas Zancudomyces was associated with non-responders. Moreover, the positive correlation interaction between Akkermansia and Rutstroemia is considered a key bacterial-fungal interaction associated with CM immunotherapy response.ConclusionOur results provide valuable insights for the enrichment of responders to immunotherapy in CM patients. Moreover, this study highlights the critical role of bacterial-fungal interactions in CM immunotherapy.

Advancements in Wind Power Forecasting: A Comprehensive Review

Accurate wind power forecasting is essential for the seamless integration of wind energy into modern power systems, yet it remains a challenging task due to the inherent variability of wind, complex atmospheric dynamics, and the influence of local terrain. Traditional forecasting models often struggle to capture these complexities, particularly for short-term and intra-hour predictions. This gap necessitates a shift towards machine learning (ML) methodologies, which have emerged as transformative tools in addressing these challenges. By leveraging large datasets and advanced algorithms, ML models can identify intricate patterns and significantly enhance prediction accuracy. Techniques such as deep learning, ensemble methods, and hybrid approaches integrate weather data with historical power output, improving both spatial and temporal resolutions. Despite their promise, challenges such as data quality, model interpretability, and computational demands require further research to fully optimize ML applications in wind power forecasting. The global transition toward smart grids, driven by the increasing penetration of renewable energy sources (RES), underscores the importance of reliable forecasting. Wind energy, as a key RES, plays a pivotal role in reducing greenhouse gas emissions and mitigating global warming. However, the stochastic nature of wind energy complicates power system analysis and management. Accurate forecasting is critical for enhancing power system security, supporting sustainability, and facilitating economic transactions in energy markets. This review examines ML-based methodologies for wind power forecasting, categorizing them into supervised, unsupervised, semi-supervised, and reinforcement learning techniques. It highlights their adaptability, scalability, and real-time capabilities while addressing challenges posed by noisy data, dynamic system behaviors, and complex grid configurations. Hybrid and ensemble models, in particular, demonstrate exceptional potential in overcoming these challenges. Furthermore, this study provides a detailed summary of the latest advancements in wind power forecasting using AI-based approaches. Key aspects such as data preparation, feature selection, and model assessment are explored, offering insights into improving both the precision and efficiency of wind energy prediction algorithms. By identifying research gaps and emerging trends, this review presents strategic directions for the development of innovative ML-driven forecasting methods. Ultimately, the findings underscore the significance of integrating advanced ML techniques to enhance forecasting reliability, support effective grid management, and contribute to a sustainable energy future.

Data-Driven Modeling and Design of Sustainable High Tg Polymers

This paper develops a machine learning methodology for the rapid and robust prediction of the glass transition temperature (Tg) for polymers for the targeted application of sustainable high-temperature polymers. The machine learning framework combines multiple techniques to develop a feature set encompassing all relative aspects of polymer chemistry, to extract and explain correlations between features and Tg, and to develop and apply a high-throughput predictive model. In this work, we identify aspects of the chemistry that most impact Tg, including a parameter related to rotational degrees of freedom and a backbone index based on a steric hindrance parameter. Building on this scientific understanding, models are developed on different types of data to ensure robustness, and experimental validation is obtained through the testing of new polymer chemistry with remarkable Tg. The ability of our model to predict Tg shows that the relevant information is contained within the topological descriptors, while the requirement of non-linear manifold transformation of the data also shows that the relationships are complex and cannot be captured through traditional regression approaches. Building on the scientific understanding obtained from the correlation analyses, coupled with the model performance, it is shown that the rigidity and interaction dynamics of the polymer structure are key to tuning for achieving targeted performance. This work has implications for future rapid optimization of chemistries

Research on Electricity Load Forecasting and Demand Side Modeling Based on the Integration of Big Data and Machine Learning

In recent times, the application of machine learning algorithms in energy forecasting has expanded significantly, facilitated by the abundant data amassed through the integration of sensor technology and power distribution systems. This data richness presents opportunities for leveraging diverse machine learning techniques. Electric load forecasting and strategic planning are pivotal for driving national industrialization and urbanization, enabling the identification of demand-side load patterns and efficient resource allocation, thereby mitigating resource waste or power shortages and fostering economic growth. This study aims to delve into the prediction and analysis of demand-side loads, leveraging the synergy of big data and machine learning methodologies. Key findings reveal that the LSTM model outperforms SVR and DNN in terms of prediction error reduction and robustness. Specifically, for single load type prediction, LSTM achieves an RMSE of 18.65% and an MAE of 18.74%. When applied to multiple load types, LSTM further enhances its performance, with RMSE and MAE declining to 9.73% and 8.89%, respectively. Aggregate load forecasting emerges as an effective strategy to minimize prediction errors stemming from individual variability, with the three models achieving RMSEs of 5.01%, 5.94%, and 6.59%, respectively, underscoring improved predictive accuracy. Notably, apparent temperature significantly influences the load prediction process, exhibiting a strong negative correlation with the average daily load.

Estimation of Gross Primary Productivity Using Performance-Optimized Machine Learning Methods for the Forest Ecosystems in China

Gross primary productivity (GPP) quantifies the rate at which plants convert atmospheric carbon dioxide into organic matter through photosynthesis, playing a vital role in the terrestrial carbon cycle. Machine learning (ML) techniques excel in handling spatiotemporally complex data, facilitating accurate spatial-scale inversion of forest GPP by integrating limited ground flux measurements with Remote Sensing (RS) observations. Enhancing ML algorithm performance for precise GPP estimation is a key research focus. This study introduces the Random Grid Search Algorithm (RGSA) for hyperparameters tuning to improve Random Forest (RF) and eXtreme Gradient Boosting (XGB) models across four major forest regions in China. Model optimization progressed through three stages: the Unoptimized (UO) XGB model achieved R2 = 0.77 and RMSE = 1.42 g Cm−2 d−1; the Hyperparameter Optimized (HO) XGB model using RGSA improved performance by 5.19% in R2 (0.81) and reduced RMSE by 9.15% (1.29 g Cm−2 d−1); the Hyperparameter and Variable Combination Optimized (HVCO) XGB model with selected variables (LAI, Temp, NR, VPD, and NDVI) further enhanced R2 to 0.83 and decreased RMSE to 1.23 g Cm−2 d−1. The optimized GPP estimates exhibited high spatial consistency with existing high-quality products like GOSIF GPP, GLASS GPP, and FLUXCOM GPP, validating the model’s reliability and effectiveness. This research provides crucial insights for improving GPP estimation accuracy and optimizing ML methodologies for forest ecosystems in China.

AI – Powered Learning Disability Detection and Classification System

Learning disabilities (LDs) are generally described as a collection of neurological disorders that interfere with an individual's ability to process and apply knowledge, affecting abilities in reading, writing, and math. To diagnose LDs properly, it is necessary to diagnose them early and accurately, though the traditional approaches used to diagnose them are slow and subjective. This paper discusses the potential of artificial intelligence, especially through machine learning and deep learning methodologies, in enhancing the detection and categorization of LDs. AI may be implemented in educational and healthcare environments to expedite and enhance the accuracy of diagnosis, which would, in turn, accelerate the results in education for people with LDs. This paper outlines the current developments in AI methods for LD identification and the potential they hold for revolutionizing diagnostic and intervention methods.

Assessment of mechanical, thermal, and sliding wear performance of chemically treated alumina‐filled biocomposites using machine learning and response surface methodology

Assessment of mechanical, thermal, and sliding wear performance of chemically treated alumina‐filled biocomposites using machine learning and response surface methodology

Unsupervised Machine Learning for Classifying CHIME Fast Radio Bursts and Investigating Empirical Relations

Abstract Fast radio bursts (FRBs) are highly energetic millisecond-duration astrophysical phenomena typically categorized as repeaters or nonrepeaters. However, observational limitations may result in misclassifications, potentially leading to a higher proportion of repeaters than currently identified. In this study, we leverage unsupervised machine learning techniques to classify FRBs using data from the CHIME/FRB catalogs, including both the first catalog and a recent repeater catalog. By employing Uniform Manifold Approximation and Projection for dimensionality reduction and clustering algorithms (k-means and Hierarchical Density-Based Spatial Clustering of Applications with Noise), we successfully segregate repeaters and nonrepeaters into distinct clusters, identifying over 100 potential repeater candidates. Our analysis reveals several empirical relations within the clusters, including the log Δ t s c − log Δ t r w , log Δ t s c − log T B , and r–γ correlations, where Δt sc , Δt rw , T B , r, and γ represent the scattering time, rest-frame width, brightness temperature, spectral running, and spectral index, respectively. The Chow test results reveal that while some repeaters and nonrepeaters share similar empirical relationships, the overall distinctions between the two groups remain significant, reinforcing the classification of FRBs into repeaters and nonrepeaters. These findings provide new insights into the physical properties and emission mechanisms of FRBs. This study demonstrates the effectiveness of unsupervised learning in classifying FRBs and identifying potential repeaters, paving the way for more precise investigations into their origins and applications in cosmology. Future improvements in observational data and machine learning methodologies are expected to further enhance our understanding of FRBs.

Machine learning for screw design in single‐screw extrusion

AbstractArtificial intelligence (AI) methods have significantly impacted various areas of technology, particularly in fields where large datasets are available. Screw designs are proprietary, and there is very limited information available in the open literature. In this study, we generated a dataset of 232 designs using computer simulation software for screw extrusion, involving solids transport, melting, and melt pumping. The parameters (features) and the outputs (targets) were introduced into four powerful machine learning (ML) algorithms. The capabilities of the four algorithms were assessed by comparing the predictions of each of the algorithms to the corresponding results of the simulations. Three of the algorithms demonstrated satisfactory performance, with the best‐performing one being further tested on an “unseen” dataset, which involved a screw of 75 mm and another of 127 mm in diameter. A machine‐learning technique called Permutation Feature Importance (PFI) was used to identify the features (parameters) with the greatest impact on the predictions. It is suggested that the same ML methodologies could be applied to datasets of existing real screw designs.Highlights Dataset obtained from simulation software. Four machine learning algorithms were employed. Assessment of algorithms based on training and testing data. Identification of parameters having greatest impact. Satisfactory predictions of mass flow rate, exit temperature, melting length, and more.

Enhancing cybersecurity through script development using machine and deep learning for advanced threat mitigation

This scholarly paper explores the utilization of Machine Learning (ML) and Deep Learning (DL) methodologies to enhance the cybersecurity aspects of script development. Given the increasing panorama of threats in contemporary software creation, cybersecurity has ascended to a critical realm of concern. Traditional security measures frequently prove inadequate in countering complex breaches. However, ML and DL present promising solutions by facilitating automated and intelligent scrutiny of security-centric tasks. In this investigation, we leverage the Fashion MNIST dataset, deploying a Convolutional Neural Network (CNN) model to underscore the efficacy of ML and DL in elevating cybersecurity. The trajectory of script development encompasses stages like data preprocessing, model training, and assessment through metrics such as accuracy and loss. Our empirical findings convincingly demonstrate that the proposed methodology yields significant enhancements in cybersecurity benchmarks, thereby validating the potential of ML and DL techniques in reinforcing software security. Furthermore, we explore practical implications and delineate the application of ML/DL integration within real software development scenarios. Through the adept amalgamation of ML and DL techniques in script development, developers can augment the robustness of their software systems against various cybersecurity threats. This paper enriches the growing body of cybersecurity research while providing invaluable insights to practitioners striving to bolster their software resilience against the ever-evolving landscape of security challenges.

Innovative Machine Learning Approaches for Drinking Water Quality Classification: Addressing Data Imbalances with Custom SMOTE Sampling Strategy

This study demonstrates the complexity and importance of water quality as a measure of the health and sustainability of ecosystems that directly influence biodiversity, human health, and the world economy. The predictability of water quality thus plays a crucial role in managing our ecosystems to make informed decisions and, hence, proper environmental management. This study addresses these challenges by proposing an effective machine learning methodology applied to the “Water Quality” public dataset. The methodology has modeled the dataset suitable for providing prediction classification analysis with high values of the evaluating parameters such as accuracy, sensitivity, and specificity. The proposed methodology is based on two novel approaches: (a) the SMOTE method to deal with unbalanced data and (b) the skillfully involved classical machine learning models. This paper uses Random Forests, Decision Trees, XGBoost, and Support Vector Machines because they can handle large datasets, train models for handling skewed datasets, and provide high accuracy in water quality classification. A key contribution of this work is the use of custom sampling strategies within the SMOTE approach, which significantly enhanced performance metrics and improved class imbalance handling. The results demonstrate significant improvements in predictive performance, achieving the highest reported metrics: accuracy (98.92% vs. 96.06%), sensitivity (98.3% vs. 71.26%), and F1 score (98.37% vs. 79.74%) using the XGBoost model. These improvements underscore the effectiveness of our custom SMOTE sampling strategies in addressing class imbalance. The findings contribute to environmental management by enabling ecology specialists to develop more accurate strategies for monitoring, assessing, and managing drinking water quality, ensuring better ecosystem and public health outcomes.

Class Imbalance Challenges in Predictive Maintenance for the Energy Power Sector: A Comprehensive Review

Imbalanced datasets, characterised by a significant disparity in class distribution, present significant obstacles in data-driven applications within the energy power sector. The presence of class imbalance can lead to the creation of biased models, decreased accuracy, and limited generalisation capabilities. In recent times, there have been several machine learning methodologies suggested to tackle the problem of class imbalance across different domains. These methodologies encompass strategies at multiple levels, such as data level, algorithm level, and the utilisation of oversampling and undersampling techniques. The objective of this study is to present an in-depth review of the current body of research pertaining to machine learning techniques employed in addressing the issue of imbalanced data within the electrical power industry. Specifically, our focus will be on evaluating several benchmarks including data preprocessing, algorithm selection, oversampling, and undersampling methodologies. This study has the potential to contribute to the advancement of more precise and dependable predictive maintenance models within the energy power industry.

RSM and AI based machine learning for quality by design development of rivaroxaban push-pull osmotic tablets and its PBPK modeling.

The study is based on applying Artificial Neural Network (ANN) based machine learning and Response Surface Methodology (RSM) as simultaneous bivariate approaches in developing controlled-release rivaroxaban (RVX) osmotic tablets. The influence of different types of polyethylene oxide, osmotic agents, coating membrane thickness, and orifice diameter on RVX release profiles was investigated. After obtaining the trial formulation data sets from Central Composite Design (CCD), an ANN-based model was trained to get the optimized formulations. The Physiological-based Pharmacokinetic (PBPK) modeling of the predicted formulation was performed by GastroPlus™ to simulate in vivo plasma profiles under fasting and fed conditions. In vitro release tests showed zero-order RVX release for up to 12h. Using graphical and numerical methods, the predicted formulation generated by the prediction profiler was cross-validated by the CCD-based optimized formulation. Analysis of Variance (ANOVA) findings revealed no significant difference between the predicted and optimized formulations and these formulations have a shelf life of 22.47 and 17.87 months, respectively. The PBPK modeling of RVX push-pull osmotic pump (PPOP) tablets suggested enhanced bioavailability in the fasted (up to 82%) and fed (up to 98.5%) state compared to immediate-release tablets. The results indicated that ANN can be effectively used for osmotic systems due to their complex nature and nonlinear interactions between dependent and independent variables.

IScore: A ML-Based Scoring Function for De Novo Drug Discovery.

In the quest for accelerating de novo drug discovery, the development of efficient and accurate scoring functions represents a fundamental challenge. This study introduces iScore, a novel machine learning (ML)-based scoring function designed to predict the binding affinity of protein-ligand complexes with remarkable speed and precision. Uniquely, iScore circumvents the conventional reliance on explicit knowledge of protein-ligand interactions and a full picture of atomic contacts, instead leveraging a set of ligand and binding pocket descriptors to directly evaluate binding affinity. This approach enables skipping the inefficient and slow conformational sampling stage, thereby enabling the rapid screening of ultrahuge molecular libraries, a crucial advancement given the practically infinite dimensions of chemical space. iScore was rigorously trained and validated using the PDBbind 2020 refined set, CASF 2016, CSAR NRC-HiQ Set1/2, DUD-E, and target fishing data sets, employing three distinct ML methodologies: Deep neural network (iScore-DNN), random forest (iScore-RF), and eXtreme gradient boosting (iScore-XGB). A hybrid model, iScore-Hybrid, was subsequently developed to incorporate the strengths of these individual base learners. The hybrid model demonstrated a Pearson correlation coefficient (R) of 0.78 and a root-mean-square error (RMSE) of 1.23 in cross-validation, outperforming the individual base learners and establishing new benchmarks for scoring power (R = 0.814, RMSE = 1.34), ranking power (ρ = 0.705), and screening power (success rate at top 10% = 73.7%). Moreover, iScore-Hybrid demonstrated great performance in the target fishing benchmarking study.

Predicting Milk Flow Behavior in Human Lactating Breast: An Integrated Machine Learning and Computational Fluid Dynamics Approach.

This study develops a comprehensive framework that integrates Computational Fluid Dynamics (CFD) and Machine Learning (ML) to predict milk flow behavior in lactating breasts. Utilizing CFD and other high-fidelity simulation techniques to tackle fluid flow challenges often entails significant computational resources and time investment. Artificial Neural Networks (ANNs) offer a promising avenue for grasping complex relationships among high-dimensional variables. This study leverages this potential to introduce an innovative data-driven approach to CFD. The initial step involved using CFD simulations to generate the necessary training and validation datasets. A machine learning pipeline was then crafted to train the ANN. Furthermore, various ANN architectures were explored, and their predictive performance was compared. The Design of Experiments (DOE) method was also harnessed to identify the minimum number of simulations needed for precise predictions. This study underscores the synergy between CFD and ML methodologies, designated as ML-CFD. This novel integration enables a neural network to generate CFD-like results, resulting in significant savings in time and computational resources typically required for traditional CFD simulations. The models developed through this ML-CFD approach demonstrate remarkable efficiency and robustness, enabling faster exploration of milk flow behavior in individual lactating breast compared to conventional CFD solvers.