Reduce Prediction Errors Research Articles

Intelligent RCS Extrapolation Technology of Target Inspired by Physical Mechanism Based on Scattering Center Model

In this paper, a technology named SCM−ANN combining physical scattering mechanisms and artificial intelligence is proposed to realize radar cross-section (RCS) extrapolation of non-cooperative conductor targets with higher efficiency. Firstly, an adaptive scattering center (SC) extraction algorithm is used to construct the scattering center model (SCM) for non-cooperative targets from radar echoes in the low-frequency band (LFB). Secondly, an artificial neural network (ANN) is constructed to capture the nonlinear relationship between the real LFB echoes and those reconstructed from the SCM. Finally, the SCM is used to reconstruct echoes in the high-frequency band (HFB), and these reconstructions, together with the trained ANN, optimize the extrapolated HFB RCS. For the SCM−ANN technology, physical mechanistic modes are used for trend prediction, and artificial intelligence is used for regression optimization based on trend prediction. Simulation results show that the proposed method can achieve a 50% frequency extrapolation range, with an average prediction error reduction of up to 40% compared with the traditional scheme. By incorporating physical mechanisms, this proposed approach offers improved accuracy and an extended extrapolation range compared with the RCS extrapolation techniques relying solely on numerical prediction.

Optimization of PV module ash accumulation prediction based on an improved vulture algorithm

Abstract In this research, an ash accumulation prediction model based on the Improved Bald Eagle Algorithm (IBES) optimization is proposed, and a least-squares support vector machine (LSSVM) is used to investigate the correlation between climatic parameters and the quantity of ash deposition on PV panel surfaces. Ash deposit affects PV panel output power over time and is influenced by meteorological factors. Following a thorough selection procedure, meteorological variables like wind speed and rainfall are used in the model to forecast the amount of ash that will accumulate on the slab. The original condor method was prone to local optimum convergence and had a slow convergence rate. These weaknesses are addressed by the optimal selection of the next iteration, which introduces the Cauchy variation operator to mutate the population’s optimal individuals. The optimization parameters of the revised algorithm are incorporated into the model and results are compared to models of other optimization algorithms. The outcomes demonstrate that the IBES-LSSVM ash accumulation prediction model has a higher degree of fit and a reduced prediction error.

Improving short-term forecasting of solar power generation by using an EEMD-BiGRU model: A comparative study based on seven standalone models and six hybrid models

ABSTRACT Accurate and timely forecasting is critical for grid-connected solar power safety and stability, achieved through machine learning (ML) for both common and real-time applications. To mitigate the impact of nonstationarity and volatility in solar power generation, we employed empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), variational mode decomposition (VMD), and complete EEMD with adaptive noise (CEEMDAN) to decompose time series into frequency components, reducing fluctuations and noise. A combination of four decomposition methods (EMD, EEMD, VMD, and CEEMDAN) and two ML models, bidirectional gated recurrent unit (BiGRU) and bidirectional long short-term memory (BiLSTM) were utilized to construct six hybrid forecasting models (EMD-BiLSTM, EMD-BiGRU, EEMD-BiLSTM, EEMD-BiGRU, VMD-BiGRU, and CEEMDAN-BiGRU), which were validated on a dataset from a 20 MW solar power station in Hebei, compared to the seven standalone ML models, backpropagation neural networks (BPNN), support vector machines (SVM), long short-term memory (LSTM), gated recurrent unit (GRU), convolutional neural networks (CNN), BiLSTM, and BiGRU, these six hybrid models demonstrated enhanced forecast accuracy. Of these, EEMD-BiGRU, VMD-BiGRU, and CEEMDAN-BiGRU significantly reduced prediction errors, with percentage reductions in root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) ranging from 44.18 ~ 49.43%, 43.67 ~ 48.59%, and 44.64% ~53.53%, respectively. The EEMD-BiGRU model outperformed all hybrid models, achieving an RMSE of 0.7662, MAE of 0.3990, MAPE of 7.982%, and R2 of 0.9865. The findings of this study can provide insights and support for applying a hybrid model based on decomposition methods for the short-term forecasting of solar power generation.

Predicting stock price of construction companies using weighted ensemble learning

Modeling the behavior of stock price data has always been one of the challenging applications of Artificial Intelligence (AI) and Machine Learning (ML) due to its high complexity and dependence on various conditions. Recent studies show that this will be difficult to do with just one learning model. The problem can be more complex for companies in the construction sector, due to the dependency of their behavior on more conditions. This study aims to provide a hybrid model for improving the accuracy of prediction for the stock price index of companies in the construction section. The contribution of this paper can be considered as follows: First, a combination of several prediction models is used to predict stock prices so that learning models can cover each other's errors. In this research, an ensemble model based on Artificial Neural Network (ANN), Gaussian Process Regression (GPR), and Classification and Regression Tree (CART) is presented for predicting the stock price index. Second, the optimization technique is used to determine the effect of each learning model on the prediction result. For this purpose, first, all three mentioned algorithms process the data simultaneously and perform the prediction operation. Then, using the Cuckoo Search (CS) algorithm, the output weight of each algorithm is determined as a coefficient. Finally, using the ensemble technique, these results are combined and the final output is generated through weighted averaging on optimal coefficients. The proposed system was implemented, and its efficiency was evaluated by real stock data of construction companies. The results showed that using CS optimization in the proposed ensemble system is highly effective in reducing prediction error. According to the results, the proposed system can predict the price index with an average accuracy of 96.6 %, which shows a reduction of at least 2.4 % in prediction error compared to the previous methods. Comparing the evaluation results of the proposed system with similar algorithms indicates that our model is more accurate and can be useful for predicting the stock price index in real-world scenarios.

Speckle Plethysmograph-Based Blood Pressure Assessment

Continuous non-invasive blood pressure (CNBP) monitoring is of the utmost importance in detecting and managing hypertension, a leading cause of death in the United States. Extensive research has delved into pioneering methods for predicting systolic and diastolic blood pressure values by leveraging pulse arrival time (PAT), the time difference between the proximal and distal signal peaks. The most widely employed pairing involves electrocardiography (ECG) and photoplethysmography (PPG). Possessing similar characteristics in terms of measuring blood flow changes, a recently investigated optical signal known as speckleplethysmography (SPG) showed its stability and high signal-to-noise ratio compared with PPG. Thus, SPG is a potential surrogate to pair with ECG for CNBP estimation. The present study aims to unlock the untapped potential of SPG as a signal for non-invasive blood pressure monitoring based on PAT. To ascertain SPG’s capabilities, eight subjects were enrolled in multiple recording sessions. A third-party device was employed for ECG and PPG measurements, while a commercial device served as the reference for arterial blood pressure (ABP). SPG measurements were obtained using a prototype smartphone-based system. Following the completion of three scenarios—sitting, walking, and running—the subjects’ signals and ABP were recorded to investigate the predictive capacity of systolic blood pressure. The collected data were processed and prepared for machine learning models, including support vector regression and decision tree regression. The models’ effectiveness was evaluated using root-mean-square error and mean absolute percentage error. In most instances, predictions utilizing PATSPG exhibited comparable or superior performance to PATPPG (i.e., SPG Rest ± 12.4 mmHg vs. PPG Rest ± 13.7 mmHg for RSME, and SPG 8% vs. PPG 9% for MAPE). Furthermore, incorporating an additional feature, namely the previous SBP value, resulted in reduced prediction errors for both signals in multiple model configurations (i.e., SPG Rest ± 12.4 mmHg to ±3.7 mmHg for RSME, and SPG Rest 8% to 3% for MAPE). These preliminary tests of SPG underscore the remarkable potential of this novel signal in PAT-based blood pressure predictions. Subsequent studies involving a larger cohort of test subjects and advancements in the SPG acquisition system hold promise for further improving the effectiveness of this newly explored signal in blood pressure monitoring.

Feature engineering of environmental covariates improves plant genomic-enabled prediction.

Because Genomic selection (GS) is a predictive methodology, it needs to guarantee high-prediction accuracies for practical implementations. However, since many factors affect the prediction performance of this methodology, its practical implementation still needs to be improved in many breeding programs. For this reason, many strategies have been explored to improve the prediction performance of this methodology. When environmental covariates are incorporated as inputs in the genomic prediction models, this information only sometimes helps increase prediction performance. For this reason, this investigation explores the use of feature engineering on the environmental covariates to enhance the prediction performance of genomic prediction models. We found that across data sets, feature engineering helps reduce prediction error regarding only the inclusion of the environmental covariates without feature engineering by 761.625% across predictors. These results are very promising regarding the potential of feature engineering to enhance prediction accuracy. However, since a significant gain in prediction accuracy was observed in only some data sets, further research is required to guarantee a robust feature engineering strategy to incorporate the environmental covariates.

Systematic review and meta-analysis of the visual mismatch negativity in schizophrenia.

Mismatch negativity (MMN) is an event-related potential component automatically elicited by events that violate predictions based on prior events. To elicit this component, researchers use stimulus repetition to induce predictions, and the MMN is obtained by subtracting the brain response to rare or unpredicted stimuli from that of frequent stimuli. Under the Predictive Processing framework, one increasingly popular interpretation of the mismatch response postulates that MMN represents a prediction error. In this context, the reduced MMN amplitude to auditory stimuli has been considered a potential biomarker of Schizophrenia, representing a reduced prediction error and the inability to update the mental model of the world based on the sensory signals. It is unclear, however, whether this amplitude reduction is specific for auditory events or if the visual MMN reveals a similar pattern in schizophrenia spectrum disorder. This review and meta-analysis aimed to summarise the available literature on the vMMN in schizophrenia. A systematic literature search resulted in 10 eligible studies that resulted in a combined effect size of g = -.63, CI [-.86, -.41], reflecting lower vMMN amplitudes in patients. These results are in line with the findings in the auditory domain. This component offers certain advantages, such as less susceptibility to overlap with components generated by attentional demands. Future studies should use vMMN to explore abnormalities in the Predictive Processing framework in different stages and groups of the SSD and increase the knowledge in the search for biomarkers in schizophrenia.

Continuous adaptive nonlinear model predictive control using spiking neural networks and real-time learning

Model predictive control (MPC) is a prominent control paradigm providing accurate state prediction and subsequent control actions for intricate dynamical systems with applications ranging from autonomous driving to star tracking. However, there is an apparent discrepancy between the model’s mathematical description and its behavior in real-world conditions, affecting its performance in real-time. In this work, we propose a novel neuromorphic (brain-inspired) spiking neural network for continuous adaptive non-linear MPC. Utilizing real-time learning, our design significantly reduces dynamic error and augments model accuracy, while simultaneously addressing unforeseen situations. We evaluated our framework using real-world scenarios in autonomous driving, implemented in a physics-driven simulation. We tested our design with various vehicles (from a Tesla Model 3 to an Ambulance) experiencing malfunctioning and swift steering scenarios. We demonstrate significant improvements in dynamic error rate compared with traditional MPC implementation with up to 89.15% median prediction error reduction with 5 spiking neurons and up to 96.08% with 5,000 neurons. Our results may pave the way for novel applications in real-time control and stimulate further studies in the adaptive control realm with spiking neural networks.

Advancing hydrogen storage predictions in metal-organic frameworks: A comparative study of LightGBM and random forest models with data enhancement

The escalating consumption of fossil fuels has given rise to a substantial upsurge in greenhouse gas concentrations and global temperatures, which, in turn, has triggered severe climate-related consequences. The critical imperative to reduce CO2 emissions and combat global warming has spurred extensive investigations into clean energy alternatives, with hydrogen emerging as a compelling zero-emission energy source. As a pivotal component of clean energy strategies, hydrogen requires designing compact, lightweight, and efficient storage systems. This study focuses on the development and evaluation of machine learning models for predicting the efficiency of Metal-Organic Frameworks (MOFs) in hydrogen storage, a key aspect of advancing clean energy technologies. MOFs, a class of nanoporous materials, show remarkable potential for hydrogen storage due to their high surface area and porosity. However, selecting the most suitable MOF for this application from a vast array of possible structures is a daunting task. In this context, machine learning algorithms offer an efficient alternative for predicting MOF suitability by considering their structural and chemical properties. We used ensemble learning methods, specifically Light Gradient Boosting Machine (LightGBM) and Random Forest (RF), to predict hydrogen uptake of MOFs based on a dataset of 219 experimentally tested samples. Two modeling scenarios were considered: one using the entire dataset, and the other involving strategic data pre-processing, including outlier removal and feature engineering. The results demonstrate that the measures taken to refine the dataset significantly enhance the predictive performance of the developed models, reducing prediction errors and improving overall goodness of fit. Specifically, the Mean Absolute Error (MAE) values for both the LightGBM and random forest models were reduced from 0.48 and 0.94, respectively, to 0.16 for both models, and the coefficients of determination (R2) increased substantially from 0.84 and 0.72 to 0.95, in both cases. Moreover, feature importance analysis unveiled that pressure-related features make the most significant contributions to the formation of tree ensembles during the model training process. A parametric sensitivity analysis was conducted revealing that H2 uptake in MOFs is most sensitive to changes in adsorption enthalpy, followed by surface area and temperature, while showing lower sensitivity to variations in pressure, consistent with established literature. These results underscore the pivotal role of data enhancement methods in refining machine learning models and can be instrumental in accelerating the development and optimization of MOF materials for clean energy applications.

Evaluation of a Microcontroller-based Smart Wearable Device in College Students' Sports Forging Application

INTRODUCTION: The widespread use of smart wearable devices in various fields, including healthcare and sports, underscores the importance of their application in enhancing physical exercise among college students. Recent advancements in technology have facilitated the development of sophisticated methods to assess and predict physical activity outcomes, making their evaluation increasingly critical.OBJECTIVES: This study aims to develop a reliable assessment model for smart wearable devices used in college students' sports activities. The objective is to accurately predict and evaluate the effectiveness of these devices in improving students' physical health and promoting lifelong sports habits. Ultimately, the research seeks to integrate advanced computational methods to enhance the accuracy of physical exercise assessments.METHODS: The research introduces a novel assessment model that combines a zebra behavior-based heuristic optimization algorithm with a convolutional neural network (CNN). By analyzing user behavior data from wearable devices, the model constructs an evaluation index system tailored for college sports activities. The approach optimizes the parameters of the CNN using the zebra optimization algorithm, ensuring enhanced prediction accuracy.RESULTS: The evaluation model demonstrated high accuracy, with a significant improvement in predicting the outcomes of physical exercises among college students. Comparative analyses with traditional methods revealed that the new model reduced prediction errors and increased real-time performance metrics. Specifically, the model achieved a lower root mean square error (RMSE) in simulation tests, indicating more precise assessments. Figures and statistical data provided in the study illustrate the model's superior performance across various parameters.CONCLUSION: The developed assessment model significantly advances the application of smart wearable devices in monitoring and enhancing college students' physical activities. By integrating cutting-edge algorithms, the study not only improves the accuracy of exercise assessments but also contributes to the broader understanding of technology's role in health and fitness education. Future research could further refine this model by incorporating additional sensors and data points to expand its applicability and robustness.

Beyond prediction error: 25 years of modeling the associations formed in the insect mushroom body.

The insect mushroom body has gained increasing attention as a system in which the computational basis of neural learning circuits can be unraveled. We now understand in detail the key locations in this circuit where synaptic associations are formed between sensory patterns and values leading to actions. However, the actual learning rule (or rules) implemented by neural activity and leading to synaptic change is still an open question. Here, I survey the diversity of answers that have been offered in computational models of this system over the past decades, including the recurring assumption-in line with top-down theories of associative learning-that the core function is to reduce prediction error. However, I will argue, a more bottom-up approach may ultimately reveal a richer algorithmic capacity in this still enigmatic brain neuropil.

Prediction of the Remaining Useful Life of Lithium-Ion Batteries Based on the 1D CNN-BLSTM Neural Network

Lithium-ion batteries are currently widely employed in a variety of applications. Precise estimation of the remaining useful life (RUL) of lithium-ion batteries holds significant function in intelligent battery management systems (BMS). Therefore, in order to increase the fidelity and stabilization of predicting the RUL of lithium-ion batteries, in this paper, an innovative strategy for RUL prediction is proposed by integrating a one-dimensional convolutional neural network (1D CNN) and a bilayer long short-term memory (BLSTM) neural network. Feature extraction is carried out through the input capacity data of the model using 1D CNN, and these deep features are used as the input of the BLSTM. The memory function of the BLSTM is applied to retain key information in the database and to better understand the coupling relationship among consecutive time series data along the time axis, thereby effectively predicting the RUL trends of lithium-ion batteries. Two different types of lithium-ion battery datasets from NASA and CALCE were used to verify the effectiveness of the proposed method. The results show that the proposed method achieves higher prediction accuracy, demonstrates stronger generalization capabilities, and effectively reduces prediction errors compared to other methods.

Towards accurate spatial prediction of Glossina pallidipes relative densities at country-scale in Kenya

Vector-borne diseases, like those transmitted by tsetse flies, pose a significant global public health threat. Reducing vector populations is a promising strategy for disease control, especially in the case of tsetse-transmitted African trypanosomiasis. However, the cost-effective implementation of large-scale vector surveillance and control measures face challenges due to the lack of spatially explicit and reliable maps identifying vector hotspots. In this study, we assessed the accuracy of predicting Glossina pallidipes relative densities across Kenya by linking constrained in-situ tsetse catch data from 660 traps across three Kenyan regions with readily available gridded satellite information (human population, land cover, soil properties, elevation, precipitation, and land surface temperature) using a classical random forest algorithm. To enhance predictive performance, we employed two feature elimination techniques specifically designed for machine learning algorithms, i.e., Recursive Feature Elimination (RFE) and Variable Selection Using Random Forests (VSURF). For each set of retained variables, we trained a Random Forest model using a spatial cross-validation technique. Our findings showed that tsetse fly relative densities decreased with mean annual precipitation, and soil moisture, and conversely increased with higher tree cover. Based on the cross-validated R2, 41% of the spatial variability in relative densities of tsetse flies could be explained. For spatial extrapolation, only the set of predictors retained by VSURF closely matched known tsetse fly distributions in Kenya. This more accurate performance of VSURF may be attributed to its approach of assessing variables for both importance and their contribution to reducing prediction error. Our study demonstrates the potential of using a random forest method to upscale tsetse relative abundance predictions to the national level. However, the reliability of the current extrapolated map remains uncertain. We recommend: 1) increasing tsetse fly sampling efforts, particularly in the data-limited northern and eastern regions of Kenya, and 2) developing a more precise and accurate land cover map with classes that directly associate with known habitat characteristics of the target tsetse species.

An effective ensemble classification framework for intrusion detection system using LSTM autoencoder

In intrusion detection, the curse of dimensionality and the trade-off between maintaining a low false alarm rate and achieving a high detection rate are significant challenges. This research suggests a unique strategy based on dimensionality reduction methods to improve the performance of network intrusion detection systems (NIDS). Compressing high-dimensional network traffic data using a Long Short-Term Memory Autoencoder (LSTMAE) allows the reduced characteristics to be submitted to a classifier to identify anomalies that may indicate an attack. Using standard datasets, including Network Security Laboratory - Knowledge Discovery in Datasets (NSL-KDD), UNSW-NB15, and Canadian Institute for Cyber Security - Intrusion Detection Systems (CICIDS2017), the proposed model is tested with classifiers like Random Forest (RF) and LightGBM (Light Gradient Boosting Machines). It is hoped that by adopting this method, NIDS response times may be improved while costs associated with storing and processing data are minimized. Precision, recall, F-score, accuracy, detection rate (DR), and false alarm rate (FAR) are only a few of the performance measures used to assess the quality of the suggested models. The experimental findings show that the proposed LSTMAE model reduces prediction errors more effectively than classic machine learning techniques such as Random Forest (RF), Gradient Boosting (GB), Support Vector Machines (SVM), Deep Belief Networks (DBN), Deep Neural Networks (DNN), Autoencoder (AE), and Long Short-Term Memory (LSTM). The results also show that the proposed solution outperforms the state-of-the-art methods of detection accuracy and computing complexity using accuracy, precision, recall, F1_Score, detection rate, and FAR.

Supplementation of deep neural networks with simplified physics-based features to increase accuracy of plate fundamental frequency predictions

To improve predictive machine learning-based models limited by sparse data, supplemental physics-related features are introduced into a deep neural network (DNN). While some approaches inject physics through differential equations or numerical simulation, improvements are possible using simplified relationships from engineering references. To evaluate this hypothesis, thin rectangular plates were simulated to generate training datasets. With plate dimensions and material properties as input features and fundamental natural frequency as the output, predictive performance of a data-driven DNN-based model is compared with models using supplemental inputs, such as modulus of rigidity. To evaluate model accuracy improvements, these additional features are injected into various DNN layers, and the network is trained with four different dataset sizes. When evaluated against independent data of similar features to the training sets, supplementation provides no statistically-significant prediction error reduction. However, notable accuracy gains occur when independent test data is of material and dimensions different from the original training set. Furthermore, when physics-enhanced data is injected into multiple DNN layers, reductions in mean error from 33.2% to 19.6%, 34.9% to 19.9%, 35.8% to 22.4%, and 43.0% to 28.4% are achieved for dataset sizes of 261, 117, 60, and 30, respectively, demonstrating potential for generalizability using a data supplementation approach. Additionally, when compared with other methods—such as linear regression and support vector machine (SVM) approaches—the physics-enhanced DNN demonstrates an order of magnitude reduction in percentage error for dataset sizes of 261, 117, and 60 and a 30% reduction for a size of 30 when compared with a cubic SVM model independently tested with data divergent from the training and validation set.

A self-attention-LSTM method for dam deformation prediction based on CEEMDAN optimization

The structural deformation of a dam directly affects its lifespan and safety, making its accurate prediction crucial. Traditional prediction methods often overlook the nonlinearity and non-smoothness of deformation data. Moreover, the irregular intervals within the historical deformation data used for model training can reduce prediction accuracy. To address these issues, we propose a hybrid deep learning model that uses signal decomposition and reconstruction to enhance dam deformation prediction. This model employs a long short-term memory (LSTM) neural network optimized using a complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). The self-attention mechanism in the LSTM model effectively captures the temporal features of dam deformation, alleviating the difficulties generated by the irregular intervals within the historical data used in model training. Furthermore, considering the lag effect of influencing factors on dam deformation and the differences among various measurement points, we propose a CEEMDAN-based feature selection method. Using 13 years worth of data from the Shuibuya Dam, we evaluate the accuracy and effectiveness of the CEEMDAN-Self-attention-LSTM model using indicators, such as MAE, RMSE, MAPE, and R2, and compared it with existing models. The experimental results show that this model reduces prediction error by more than 53.62%.

Алгоритм COOT Bird для ежедневного прогнозирования концентрации мелкодисперсных твердых частиц. Статистическое исследование

Fine particulate matter (PM2.5) poses significant risks to public health and the natural environment. Accurate prediction of PM2.5 concentration is crucial for effective environmental management. In this study, we present a novel hybrid model, the COOT bird-inspired natural life model combined with Artificial Neural Network (COOT-ANN), for predicting daily PM2.5 concentration in hydier abad and Delhi from 2014 to 2022. The performance of the COOT-ANN model is compared with stand-alone ANN and Dragonfly-ANN (DA-ANN) hybrid models. Using the Taylor diagram, we demonstrate that the COOT-ANN model exhibits the closest proximity to the observation point, resulting in a 13.94 % and 11.42 % reduction in prediction errors compared to the ANN model in Hyderabad and Delhi, respectively. Furthermore, the box-plot of the COOT-ANN model closely resembles the actual data distribution. Consequently, the COOT-ANN model outperforms both the ANN and DA-ANN models at both monitoring stations. This innovative approach to air quality prediction can significantly enhance the accuracy of environmental protection programs.

Chemical space deconstruction-based dynamic model ensemble architecture for molecular property prediction

Green solvents, catalysts, functional materials, drugs, and other chemical products now have a quick engine for design thanks to machine learning (ML) based prediction of molecular properties. However, the accuracy and stability of ML-based models can be impeded by poor data quality, which is rarely studied in chemical product discovery and design. Inspired by the dynamic ensemble selection (DES), an improved DES based on chemical space deconstruction is proposed in this work to accommodate the prediction task of molecular properties. We innovatively developed a chemical space representation and deconstruction model based on a self-organizing mapping (SOM) neural network, facilitating the rapid implementation of the improved DES on molecular samples. Consequently, a novel dynamic model ensemble architecture (SOM-DES) is proposed as a model enhancement technology to build a more accurate and stable ensemble model, aiming to improve the predictive performance on the chemical subspace within poor-quality data. To achieve the architecture, a supervised dimensionality reduction algorithm has been improved to enhance the deep mining of molecular feature information for DES optimization. Additionally, a novel resampling strategy based on the combination of the geometric synthetic minority oversampling technique (G-SMOTE) algorithm and chemical space deconstruction, as a data augmentation technology, has been proposed for mitigating the disadvantage of unbalanced data during DES training. The prediction task for enthalpy of formation of the ideal gas is applied as a case study to demonstrate the superiority of the proposed SOM-DES. The results indicate that the proposed SOM-DES (R2 = 0.9731, RMSE = 55.4639) outperforms the traditional static ensemble strategy (SES, R2 = 0.9552, RMSE = 71.5045) in terms of global chemical spatial precision. More importantly, for chemical subspaces that are difficult to predict due to low data quality, SOM-DES shows a significant reduction in prediction errors compared to SES.

Data-driven prediction of aerodynamic noise of transonic buffeting over an airfoil

Accurately predicting buffet frequency and aerodynamic noise level is crucial in transonic buffet noise reduction studies. In this study, the Random Forest (RF) algorithm is employed to predict the Power Spectral Density (PSD) and Overall Sound Pressure Level (OASPL) distribution over the supercritical airfoil RAE2822. The study indicates that the RF algorithm exhibits greater advantages over the Multi-Layer Perceptron (MLP). This algorithm does not suffer from the problems of high-frequency divergence or inflection distortion, maintaining a reduction in prediction errors for OASPL and PSD by approximately three to four orders of magnitude. Additionally, this paper proposes a priori criterion for evaluating the accuracy of PSD prediction. When there is a strong correlation between PSD at adjacent points, a data-driven modeling approach can achieve higher prediction accuracy. However, when the root mean square error of the cross-correlation functions, auto-correlation functions, and the PSD between adjacent points are both high, the generalization ability of pure data-driven modeling is insufficient, necessitating the additional monitoring points to ensure prediction accuracy.

Machine learning accelerates pharmacophore-based virtual screening of MAO inhibitors

Nowadays, an efficient and robust virtual screening procedure is crucial in the drug discovery process, especially when performed on large and chemically diverse databases. Virtual screening methods, like molecular docking and classic QSAR models, are limited in their ability to handle vast numbers of compounds and to learn from scarce data, respectively. In this study, we introduce a universal methodology that uses a machine learning-based approach to predict docking scores without the need for time-consuming molecular docking procedures. The developed protocol yielded 1000 times faster binding energy predictions than classical docking-based screening. The proposed predictive model learns from docking results, allowing users to choose their preferred docking software without relying on insufficient and incoherent experimental activity data. The methodology described employs multiple types of molecular fingerprints and descriptors to construct an ensemble model that further reduces prediction errors and is capable of delivering highly precise docking score values for monoamine oxidase ligands, enabling faster identification of promising compounds. An extensive pharmacophore-constrained screening of the ZINC database resulted in a selection of 24 compounds that were synthesized and evaluated for their biological activity. A preliminary screen discovered weak inhibitors of MAO-A with a percentage efficiency index close to a known drug at the lowest tested concentration. The approach presented here can be successfully applied to other biological targets as target-specific knowledge is not incorporated at the screening phase.