Linear Learning Algorithm Research Articles

Decoding wheat contamination through self-assembled whole-cell biosensor combined with linear and non-linear machine learning algorithms

The contamination of mycotoxins is a serious problem around the world. It has detrimental effects on human beings and leads to tremendous economic loss. It is essential to develop a rapid and non-destructive method for contamination recognition particularly for early alarm. In this study, the whole-cell biosensor array was constructed and employed for rapid recognition of wheat contamination by combining with machine learning algorithms. Seven key VOCs were explored through univariate coupling to multivariate analysis of orthogonal partial least squares-discrimination analysis (OPLS-DA) models. The promoters of dnaK, katG, oxyR, soxS obtained from the stress-responsive of key VOCs were fused to the bacterial operon and fabricated on the whole-cell biosensor. The constructed whole-cell biosensor array was consisted with four kinds of sensors and 18 sensor unit. The bioluminescent intensity combined with linear machine learning algorithm of partial least squares discriminant analysis (PLS-DA) and non-linear algorithms of back propagating artificial neural network (BP-ANN) and least square support vector machine (LS-SVM) were employed to establish discrimination models for mold contamination especially for early warning. The Monte-Carlo strategy was performed to generate thirty subsets for modeling to give more reliable results. As a result, the whole-cell biosensor combined with non-linear algorithm of LS-SVM was practicable for detecting mold identification for wheat early-warning with the accuracy of 97.24%. Additionally, this study provides practical and effective methods not only for wheat quality guarantee and supervision but also for other foodstuffs.

Machine learning analysis/optimization of auxetic performance of a polymeric meta-hybrid structure of re-entrant and meta-trichiral

In the ever-evolving field of materials science, the quest for innovative design solutions continues. One such area of research is the optimization of auxetic performance in meta-hybrid structures. This study delves into the application of machine learning analysis to enhance the properties of a specific meta-hybrid structure combining re-entrant and meta-trichiral designs. The polymeric meta-hybrid structure is subjected to finite element (FE) analysis to determine its Poisson's ratio. Various machine learning methods, such as linear, quadratic, and non-linear approaches, are employed to accurate forecast the Poisson's ratio of the polymeric meta-hybrid structure. Although both linear and quadratic machine learning algorithms have shown some success in predicting Poisson's ratios, they cannot be considered outstanding. As a result, a non-linear machine learning algorithm has been developed to enhance the accuracy of Poisson's ratio prediction. The results indicate that non-linear machine learning techniques out performs both linear and quadratic methods when it came to predicting Poisson's ratio. Specifically, the non-linear model achieved an impressive R-sq value of 93.68%, demonstrating its exceptional accuracy and overall excellence in fitting the data. The optimum conditions for the meta-hybrid structure are determined to be a thickness of 4.35 mm, a temperature of 15 °C, and a strain rate of 1%. Under these specific parameters, the Poisson's ratio of the meta-hybrid structure is measured to be −3.899, well outside isotropic stability limit but acceptable for auxetic structures.

Performance analysis and optimization of SnSe thin-film solar cell with Cu2O HTL through a combination of SCAPS-1D and machine learning approaches

Tin selenide (SnSe) is an auspicious light-harvesting material in photovoltaic (PV) technology due to its larger absorption coefficients and earth-abundance nature. However, because of the potential difficulty of defect-free fabrication, and non-optimized electron transport layer (ETL) and hole transport layer (HTL) alignment, it could no longer achieve its realistic objectives. Therefore, designing a suitable SnSe-based PV device with an appropriate ETL and HTL is crucial to achieving optimum performance. In this research, we proposed a novel heterojunction thin-film solar cell (TFSC) configuration of Ni/Cu2O/SnSe/WS2/FTO/Al and simulated its PV performance metrics using SCAPS-1D solar cell simulation software. This research additionally provided a comparative performance analysis of the SnSe TFSC with numerous ETLs and HTLs. It is evident that the suggested TFSC with WS2 ETL and Cu2O HTL creates appropriate band alignment with the SnSe light harvesting layer, which reduces charge recombination at both the ETL/absorber and absorber/HTL interfaces. Here, the impact of absorber layer thickness, doping concentration, bulk defect density, and defect density at the ETL/absorber and absorber/HTL interfaces is investigated. Besides, the impact of radiative recombination, back contact metal work function, and working temperature are carefully examined. After optimization of all the material properties, a maximum efficiency of 29.31 % with open circuit voltage (Voc) of 0.89 V, short circuit current density (Jsc) of 38.23 mA/cm2, and fill-factor (FF) of 86.30 % were attained for the SnSe-based proposed structure. Furthermore, a supervised linear regression machine learning algorithm is employed to investigate how the behavior of the material attributes affects the solar cell's designed PV performance. Overall, our findings can be helpful to experimentalists to fabricate inexpensive, highly efficient, and Cd-free SnSe-based heterostructure solar cells in the near future.

Prediction of primary admission total charges following cervical disc arthroplasty utilizing machine learning

BACKGROUND CONTEXTCervical disc arthroplasty (CDA) has become an increasingly popular alternative to anterior cervical discectomy and fusion, offering benefits such as motion preservation and reduced risk of adjacent segment disease. Despite its advantages, understanding the economic implications associated with varying patient and hospital factors remains critical. PURPOSETo evaluate how hospital size, geographic region, and patient-specific variables influence charges associated with the primary admission period following CDA. STUDY DESIGNA retrospective analysis using machine learning models to predict and analyze charge factors associated with CDA. PATIENT SAMPLEData from the National Inpatient Sample (NIS) database from 2016 to 2020 was used, focusing on patients undergoing CDA. OUTCOME MEASURESThe primary outcome was total charge associated with the primary admission for CDA, analyzed against patient demographics, hospital characteristics, and regional economic conditions. METHODSMultivariate linear regression and machine learning algorithms including logistic regression, random forest, and gradient boosting trees were employed to assess their predictive power on charge outcomes. Statistical significance was set at the 0.003 level after applying a Bonferroni correction. RESULTSThe analysis included 3,772 eligible CDA cases. Major predictors of charge identified were hospital size and ownership type, with large and privately owned hospitals associated with higher charges (p<.001). The Western region of the U.S. also showed significantly higher charges compared to the Northeast (p<.001). The gradient boosting trees model showed the highest accuracy (AUC=85.6%). Length of stay and wage index were significant charge drivers, with each additional inpatient day increasing charges significantly (p<.001) and higher wage index regions correlating with increased charges (p<.001). CONCLUSIONSHospital size, geographic region, and specific patient demographics significantly influence the charges of CDA. Machine learning models proved effective in predicting these charges, suggesting that they could be instrumental in guiding economic decision-making in spine surgery. Future efforts should aim to incorporate these models into broader clinical practice to optimize healthcare spending and enhance patient care outcomes.

Comparison of machine learning algorithms and multiple linear regression for live weight estimation of Akkaraman lambs

This study was designed to predict the post-weaning weights of Akkaraman lambs reared on different farms using multiple linear regression and machine learning algorithms. The effect of factors the age of the dam, gender, type of lambing, enterprise, type of flock, birth weight, and weaning weight was analyzed. The data was collected from a total of 25,316 Akkaraman lambs raised at multiple farms in the Çiftlik District of Niğde province. Comparative analysis was conducted by using multiple linear regression, Random Forest, Support Vector Machines (and Support Vector Regression), Extreme Gradient Boosting (XGBoost) (and Gradient Boosting), Bayesian Regularized Neural Network, Radial Basis Function Neural Network, Classification and Regression Trees, Exhaustive Chi-squared Automatic Interaction Detection (and Chi-squared Automatic Interaction Detection), and Multivariate Adaptive Regression Splines algorithms. In this study, the test dataset was divided into five layers using the K-fold cross-validation method. The performance of models was compared using performance criteria such as Adjusted R-squared (Adj-\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${R}^{2}$$\\end{document}), Root Mean Square Error (RMSE), Mean Absolute Deviation (MAD), and Mean Absolute Percentage Error (MAPE) by utilizing test populations in the predicted models. Additionally, the presence of low standard deviations for these criteria indicates the absence of an overfitting problem. \\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${R}^{2}$$\\end{document}The comparison results showed the Random Forest algorithm had the best predictive performance compared to other algorithms with Adj-\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${R}^{2}$$\\end{document}, RMSE, MAD, and MAPE values of 0.75, 3.683, 2.876, and 10.112, respectively. In conclusion, the results obtained through Multiple Linear Regression for the live weights of Akkaraman lambs were less accurate than the results obtained through artificial neural network analysis.

Heuristics for the Intelligent Prediction of Population Growth

Population growth is a phenomenon that is inevitable in a life course and the components of population growth are fertility, mortality and migration. Because of the impact population growth exact upon the socio- economic nature of a country, it will be wise to know the future size and distribution of it so that adequate measures can be made to forestall possible problems. There are different population models and algorithms implemented to project population growth, ranging from statistical to machine learning models; these model techniques were very suitable in their time, such as the Malthusian theory, which, first and foremost, gives deeper consideration to the exponential growth model, while many implemented classification and linear regression algorithm. Linear regression machine learning algorithms were considered the most effective algorithm for population growth projection. This study tends to develop a machine learning model that is data-driven mathematically, capable of implementing the data-independent prediction model equation that was used to predict the impact of the Non-pharmaceuticals approach and pharmaceuticals approach (NPA and PA) in mitigating the spread of the Covid-19 virus, and both models were re-modified to project Nigeria population growth. The data-independent prediction model (DIPM) utilized onset data from the NPA and PA interventions to predict the probability of mitigating the spread of the virus for a specific period. The DIPM has the properties of arithmetic and exponential methods. The fusion of these two properties with the aid of machine learning model has further reveal the data-independent prediction model as a conceptualization technique to reflect how data are processed in an algorithm setting in a concrete world, with the support of the Java array list algorithm, what all the statistical models and supervised machines learning model used in the past studies could not achieve, have been accomplished succinctly. The data independent prediction model is a robust technique for both projection and forecasting future population growth as well as proffer answers to historical issues in mathematical modeling of population expansion.

Combining algorithm techniques with mechanical and acoustic profiles for the prediction of apples sensory attributes

The research work shows the potentiality of advanced linear and nonlinear learning algorithm techniques in the prediction of apples texture sensory attributes as “hardness”, “crunchiness”, “flouriness”, “fibrousness”, and “graininess”. Starting from the information contained in the entire mechanical and acoustic curves acquired during samples compression test, the prediction performances of five different statistical tools as Partial Least Squares regression (PLS), Multilayer Perceptron (MLP), Support Vector Regression (SVR) and Gaussian Process Regression (GPR) are shown and discussed.All Predictive models validations evidence best accuracies for texture sensory attributes “hardness” and “crunchiness” and in general for GPR learning algorithm. By combining mechanical and acoustic profiles, 5-fold cross validations produce values of coefficient of determination R2 up to 0.885 (GPR) and 0.840 (GPR), respectively for “hardness” and “crunchiness”. These results, comparable to those obtained by considering a large number of mechanical and acoustic parameters extracted from acquired profiles as predictive factors, evidence a new and reliable way for the prediction of texture sensory attributes of apples. The proposed approach can overcome the necessity to define, in advance, number and type of features to be calculated from instrumental texture profiles and can be easily implemented in an automatic process.

A new technique for handling non-probability samples based on model-assisted kernel weighting

Surveys are going through massive changes, and the most important innovation is the use of non-probability samples. Non-probability samples are increasingly used for their low research costs and the speed of the attainment of results, but these surveys are expected to have strong selection bias caused by several mechanisms that can eventually lead to unreliable estimates of the population parameters of interest. Thus, the classical methods of statistical inference do not apply because the probabilities of inclusion in the sample for individual members of the population are not known. Therefore, in the last few decades, new possibilities of inference from non-probability sources have appeared.Statistical theory offers different methods for addressing selection bias based on the availability of auxiliary information about other variables related to the main variable, which must have been measured in the non-probability sample. Two important approaches are inverse probability weighting and mass imputation. Other methods can be regarded as combinations of these two approaches.This study proposes a new estimation technique for non-probability samples. We call this technique model-assisted kernel weighting, which is combined with some machine learning techniques. The proposed technique is evaluated in a simulation study using data from a population and drawing samples using designs with varying levels of complexity for, a study on the relative bias and mean squared error in this estimator under certain conditions. After analyzing the results, we see that the proposed estimator has the smallest value of both the relative bias and the mean squared error when considering different sample sizes, and in general, the kernel weighting methods reduced more bias compared to based on inverse weighting. We also studied the behavior of the estimators using different techniques such us generalized linear regression versus machine learning algorithms, but we have not been able to find a method that is the best in all cases. Finally, we study the influence of the density function used, triangular or standard normal functions, and conclude that they work similarly.A case study involving a non-probability sample that took place during the COVID-19 lockdown was conducted to verify the real performance of the proposed methodology, obtain a better estimate, and control the value of the variance.

Development and Validation of Prediction Models for Incident Reversible Cognitive Frailty Based on Social-Ecological Predictors Using Generalized Linear Mixed Model and Machine Learning Algorithms: A Prospective Cohort Study.

This study aimed to develop and validate prediction models for incident reversible cognitive frailty (RCF) based on social-ecological predictors. Older adults aged ≥60 years from China Health and Retirement Longitudinal Study (CHARLS) 2011-2013 survey were included as training set (n = 1230). The generalized linear mixed model (GLMM), eXtreme Gradient Boosting, support vector machine, random forest, and Binary Mixed Model forest were used to develop prediction models. All models were evaluated internally with 5-fold cross-validation and evaluated externally via CHARLS 2013-2015 survey (n = 1631). Only GLMM showed good discrimination (AUC = 0.765, 95% CI = 0.736, 0.795) in training set, and all models showed fair discrimination (AUC = 0.578-0.667, 95% CI = 0.545, 0.725) in internal and external validation. All models showed acceptable calibration, overall prediction performance, and clinical usefulness in training and validation sets. Older adults were divided into three groups using risk score based on GLMM, which could assist healthcare providers to predict incident RCF, facilitating early identification of high-risk population.

Artificial Intelligence Machine Learning Algorithms Versus Standard Linear Demographic Analysis in Predicting Implant Size of Anatomic and Reverse Total Shoulder Arthroplasty.

Accurate and precise templating is paramount for anatomic total shoulder arthroplasty (TSA) and reverse total shoulder arthroplasty (RSA) to enhance preoperative planning, streamline surgery, and improve implant positioning. We aimed to evaluate the predictive potential of readily available patient demographic data in TSA and RSA implant sizing, independent of implant design. A total of 578 consecutive, primary, noncemented shoulder arthroplasty cases were retrospectively reviewed. Demographic variables and implant characteristics were recorded. Multivariate linear regressions were conducted to predict implant sizes using patient demographic variables. Linear models accurately predict TSA implant sizes within 2 millimeters of humerus stem sizes 75.3% of the time, head diameter 82.1%, head height 82.1%, and RSA glenosphere diameter 77.6% of the time. Linear models predict glenoid implant sizes accurately 68.2% and polyethylene thickness 76.6% of the time and within one size 100% and 95.7% of the time, respectively. Linear models accurately predict shoulder arthroplasty implant sizes from demographic data. No significant statistical differences were observed between linear models and machine learning algorithms, although the analysis was underpowered. Future sufficiently powered studies are required for more robust assessment of machine learning models in predicting primary shoulder arthroplasty implant sizes based on patient demographics.

Catalytic Activity of 2-Imino-1,10-phenthrolyl Fe/Co Complexes via Linear Machine Learning.

In anticipation of the correlations between catalyst structures and their properties, the catalytic activities of 2-imino-1,10-phenanthrolyl iron and cobalt metal complexes are quantitatively investigated via linear machine learning (ML) algorithms. Comparatively, the Ridge Regression (RR) model has captured more robust predictive performance compared with other linear algorithms, with a correlation coefficient value of R2= 0.952 and a cross-validation value of Q2= 0.871. It shows that different algorithms select distinct types of descriptors, depending on the importance of descriptors. Through the interpretation of the RR model, the catalytic activity is potentially related to the steric effect of substituents and negative charged groups. This study refines descriptor selection for accurate modeling, providing insights into the variation principle of catalytic activity.

Linear discriminant analysis-based deep learning algorithms for numerical character handwriting recognition

Linear discriminant analysis-based deep learning algorithms for numerical character handwriting recognition

Sex determination based on craniometric parameters: a comparative approach between linear and non-linear machine learning algorithms

Introduction: Determining sex based on cranial characteristics is of great relevance in forensic anthropology. Most studies have employed linear methods (such as logistic regression) for this estimation with accuracies around 70%, rarely exceeding 90% accuracy. Several authors have tested non-linear models such as neural networks, support vector machines, and decision trees with good results, surpassing linear models. Objective: To compare linear models (logistic regression, linear regression, and linear discriminant analysis) with non-linear models (neural networks, extreme gradient boosting, support vector machine, naive Bayes, random forest, decision tree, k-nearest neighbors, and adaptive multivariate spline regression). Materials and Methods: 241 skulls used in this study were obtained from the collection of Center for Study and Research in Anatomy and Forensic Anthropology at Tiradentes University, Farolândia campus in Aracaju, Sergipe. Each skull in the collection has secure detailed records. Eighty-nine skulls with signs of craniotomy (n=58) or damage (n=30) and one unidentified were excluded. The 152 eligible skulls underwent cranial measurements. Using the Anaconda platform and Jupyter editor, the data were divided into a training group (80% of the sample) and then were tested (20% of the sample). Eleven machine learning algorithms, including both linear and non-linear models, were applied. Results: The best machine learning algorithm was a neural network with average accuracy of 93%, after 50 runs. The difference to logistic regression, which had an accuracy of 68%, was significantly (p-value of 0.01016). Conclusion: This study demonstrated the potential of the neural network for solving the sex classification problem. The study has a limitation in that neural networks perform better with a large volume of data, and this study used data from a single center. Nevertheless, in the future, more studies should be conducted testing neural networks with larger samples and skulls from other continents.

Estimating the aboveground biomass of the Hulunbuir Grassland and exploring its spatial and temporal variations over the past ten years

In the past 10 years, extreme weather phenomena have increased, and global warming has markedly advanced; moreover, the intensity of human activity has gradually increased. These have an impact on the growth of global vegetation. Related studies have focused on the Tibetan Plateau and some northern provinces of China to estimate grassland AGB at a large scale but a low resolution. Hulunbuir Grassland is an important supplier of livestock products, and therefore, it is important to precisely map AGB and explore the response of Hulunbuir Grassland AGB to climate change and human activities at a high resolution to identify complex spatial details. In this study, we selected vegetation indices from Landsat 8 OLI and topographic indices and used multiple linear regression and machine learning algorithms to estimate the spatial distribution of AGB from 2013 to 2022. Then, we analyzed the correlations between AGB and cumulative precipitation and daily average temperature in summer and between population density and livestock density at the pixel level. Our results demonstrated that the RF model performed well, with an RMSE of 28.23 and R2 value of 0.74; the AGB was positively correlated with the cumulative precipitation in summer in 94.45 % of the area and negatively correlated with the daily average temperature in summer in 96.32 % of the area. We suggest that it is necessary to reduce grazing activities in future warm and drought years and to adjust sources of income to adapt to the decrease in AGB under global warming conditions. This study will provide reference for the countries or regions that depend on temperate grasslands for husbandry.

Machine learning approach to predict the turbidity of Saki Lake, Telangana, India, using remote sensing data

Water quality is crucial for all life forms, yet water pollution is escalating. Monitoring water quality is essential to combat this challenge. This study introduces a precise and efficient approach to predict water turbidity levels using linear regression models and machine learning algorithms such as k-NN regression and decision trees. The model is trained using independent features like red band reflectance and NDTI. Hyperparameter tuning, utilizing grid search CV and repeated k-fold cross-validation, is applied to enhance the model's accuracy. The machine learning method was assessed with turbidity data measured from Saki Lake in Hyderabad, India, over four years (2014–2017) by the Telangana State Groundwater Department. Concurrently, Landsat-8 imagery from the USGS was employed for comprehensive analysis. The decision tree regression, optimized with hyperparameter tuning, outperformed the others, yielding an MAE of 3.246, an RMSE of 3.802, and a correlation coefficient (R2) of 0.776. This study validates the decision tree method's precision in forecasting water turbidity and its strong agreement with on-site measured values.

Influence of psychological and personal factors on predicting individual’s thermal comfort in an office building using linear estimation and machine learning model

ABSTRACT Thermal comfort in the building affects occupants’ health, productivity, and electricity use. Predicting occupants’ thermal comfort in advance will be helpful. Nowadays, a widely used model for thermal comfort prediction is the predicted mean vote (PMV). However, studies have found discrepancies between PMV and thermal sensation votes. In this paper, a comparative study was made by developing thermal comfort prediction in office building models using linear estimation and machine learning (ML) algorithms. Fanger’s six parameters and nine other parameters are considered for linear estimation and ML input parameters. These parameters are divided into two groups: the psychology group with the parameter’s thermal preference, thermal acceptance, overall comfort, air movement vote, humidity preference, and humidity vote, and the personal group with the parameters of gender and age. The predictive model developed showed the validated coefficient of correlation of more than 0.88 for both categories. For ML, training and evaluation have been done for five ML models: Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and K-Nearest Neighbour (KNN). The findings showed that new parameters made significantly better thermal comfort prediction. RF has the lowest prediction average error, 0.706 when including all the psychological group parameters.

Modeling claims frequency in the Algerian automobile insurance market using machine learning

The Algerian automobile insurance market faces significant challenges in pricing insurance policies due to the lack of reliable predictions for insurance losses. In this paper, we introduce a new ratemaking system that leverages advanced data analysis techniques, including Generalized Linear Models and machine learning algorithms like Neural Networks, boosting, and stacking algorithms, to model claims frequency. By analyzing data and statistics of drivers in the Algerian market, this system offers a data-driven solution that helps insurers to better understand their risk exposure and make informed pricing decisions. The proposed system has implications for both insurers and policyholders in terms of fairer and more accurate pricing, which will ultimately benefit the Algerian economy.&#x0D;

A hybrid model for data visualization using linear algebra methods and machine learning algorithm

The t-distributed stochastic neighbor embedding (t-SNE) is a powerful technique for visualizing high-dimensional datasets. By reducing the dimensionality of the data, t-SNE transforms it into a format that can be more easily understood and analyzed. The existing approach is to visualize high-dimensional data but not deeply visualize. This paper proposes a model that enhances visualization and improves the accuracy. The proposed model combines the non-linear embedding technique t-SNE, the linear dimensionality reduction method principal component analysis (PCA), and the QR decomposition algorithm for discovering eigenvalues and eigenvectors. In Addition, we quantitatively compare the proposed model QRPCA-t-SNE with PCA-t-SNE using the following criteria: data visualization with different perplexity and different principal components, confusion matrix, model score, mean square error (MSE), training, testing accuracy, receiver operating characteristic curve (ROC) score, and AUC score.

Cybersecurity and Risk Prediction Based on Machine Learning Algorithms

Abstract This paper focuses on the application of machine learning in cybersecurity and risk prediction. Considering the complexity and dynamics of the cyber environment, the role of plain Bayesian algorithms in improving the accuracy of cybersecurity prediction is explored. First, the limitations of traditional linear regression machine learning algorithms in cybersecurity are analyzed, pointing out that it is difficult for these algorithms to capture complex nonlinear relationships. For this reason, plain Bayes is introduced to improve machine learning for more accurate prediction and assessment of cybersecurity risks by capturing functions and probabilistic agent models. In terms of methodology, this paper combines various technical tools such as statistical analysis, feature extraction, and risk assessment. The research results show that the improved algorithm outperforms the traditional method in simulation experiments, which is reflected in the 10% improvement in risk assessment accuracy and 20% enhancement of detection sensitivity to network intrusion. In addition, through big data analysis, the algorithm successfully identifies key risk points in network security and effectively predicts potential security threats. The improved machine learning algorithm with plain Bayes has significant advantages in network security and risk prediction and can effectively enhance network security protection capability and prediction accuracy.

Scaling behaviours of deep learning and linear algorithms for the prediction of stroke severity.

Deep learning has allowed for remarkable progress in many medical scenarios. Deep learning prediction models often require 105-107 examples. It is currently unknown whether deep learning can also enhance predictions of symptoms post-stroke in real-world samples of stroke patients that are often several magnitudes smaller. Such stroke outcome predictions however could be particularly instrumental in guiding acute clinical and rehabilitation care decisions. We here compared the capacities of classically used linear and novel deep learning algorithms in their prediction of stroke severity. Our analyses relied on a total of 1430 patients assembled from the MRI-Genetics Interface Exploration collaboration and a Massachusetts General Hospital-based study. The outcome of interest was National Institutes of Health Stroke Scale-based stroke severity in the acute phase after ischaemic stroke onset, which we predict by means of MRI-derived lesion location. We automatically derived lesion segmentations from diffusion-weighted clinical MRI scans, performed spatial normalization and included a principal component analysis step, retaining 95% of the variance of the original data. We then repeatedly separated a train, validation and test set to investigate the effects of sample size; we subsampled the train set to 100, 300 and 900 and trained the algorithms to predict the stroke severity score for each sample size with regularized linear regression and an eight-layered neural network. We selected hyperparameters on the validation set. We evaluated model performance based on the explained variance (R2) in the test set. While linear regression performed significantly better for a sample size of 100 patients, deep learning started to significantly outperform linear regression when trained on 900 patients. Average prediction performance improved by ∼20% when increasing the sample size 9× [maximum for 100 patients: 0.279 ± 0.005 (R2, 95% confidence interval), 900 patients: 0.337 ± 0.006]. In summary, for sample sizes of 900 patients, deep learning showed a higher prediction performance than typically employed linear methods. These findings suggest the existence of non-linear relationships between lesion location and stroke severity that can be utilized for an improved prediction performance for larger sample sizes.