Linear Machine Learning Research Articles

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.

Predicting the individual effects of team competition on college students’ academic performance in mobile edge computing

Mobile edge computing (MEC) has revolutionized the way of teaching in universities. It enables more interactive and immersive experiences in the classroom, enhancing student engagement and learning outcomes. As an incentive mechanism based on social identity and contest theories, team competition has been adopted and shown its effectiveness in improving students’ participation and motivation in college classrooms. However, despite its potential benefit, there are still many unresolved issues: What type of students and teams benefit more from team competition? In what teaching context is team competition more effective? Which competition design methods better increase student academic performance? Mobile edge computing provides the ability to obtain the data of the teaching process and analyze the causal effect between team competition and students’ academic performance. In this paper, the authors first design a randomized field experiment among freshmen enrolled in college English courses. Then, the authors analyze the observation data collected from the online teaching platform, and predict individual treatment effects of academic performance in college English through linear and nonlinear machine learning models. Finally, by carefully investigating features of teams and individual students, the prediction error is reduced by up to 30%. In addition, through interpreting the predictive models, some valuable insights regarding the practice of team competition in college classrooms are discovered.

How close are the classical two-body potentials to abinitio calculations? Insights from linear machine learning based force matching.

In this work, we propose a linear machine learning force matching approach that can directly extract pair atomic interactions from abinitio calculations in amorphous structures. The local feature representation is specifically chosen to make the linear weights a force field as a force/potential function of the atom pair distance. Consequently, this set of functions is the closest representation of the abinitio forces, given the two-body approximation and finite scanning in the configurational space. We validate this approach in amorphous silica. Potentials in the new force field (consisting of tabulated Si-Si, Si-O, and O-O potentials) are significantly different than existing potentials that are commonly used for silica, even though all of them produce the tetrahedral network structure and roughly similar glass properties. This suggests that the commonly used classical force fields do not offer fundamentally accurate representations of the atomic interaction in silica. The new force field furthermore produces a lower glass transition temperature (Tg ∼ 1800K) and a positive liquid thermal expansion coefficient, suggesting the extraordinarily high Tg and negative liquid thermal expansion of simulated silica could be artifacts of previously developed classical potentials. Overall, the proposed approach provides a fundamental yet intuitive way to evaluate two-body potentials against abinitio calculations, thereby offering an efficient way to guide the development of classical force fields.

A Generalized Linear Model and Machine Learning Approach for Predicting the Frequency and Severity of Cargo Insurance in Thailand’s Border Trade Context

The study compares model approaches in predictive modeling for claim frequency and severity within the cross-border cargo insurance domain. The aim is to identify the optimal model approach between generalized linear models (GLMs) and advanced machine learning techniques. Evaluations focus on mean absolute error (MAE) and root mean squared error (RMSE) metrics to comprehensively assess predictive performance. For frequency prediction, extreme gradient boosting (XGBoost) demonstrates the lowest MAE, indicating higher accuracy compared to gradient boosting machines (GBMs) and a generalized linear model (Poisson). Despite XGBoost’s lower MAE, it shows higher RMSE values, suggesting a broader error spread and larger magnitudes compared to gradient boosting machines (GBMs) and a generalized linear model (Poisson). Conversely, the generalized linear model (Poisson) showcases the best RMSE values, indicating tighter clustering and smaller error magnitudes, despite a slightly higher MAE. For severity prediction, extreme gradient boosting (XGBoost) displays the lowest MAE, implying better accuracy. However, it exhibits a higher RMSE, indicating wider error dispersion compared to a generalized linear model (Gamma). In contrast, a generalized linear model (Gamma) demonstrates the lowest RMSE, portraying tighter clustering and smaller error magnitudes despite a higher MAE. In conclusion, extreme gradient boosting (XGBoost) stands out in mean absolute error (MAE) for both frequency and severity prediction, showcasing superior accuracy. However, a generalized linear model (Gamma) offers a balance between accuracy and error magnitude, and its performance outperforms extreme gradient boosting (XGBoost) and gradient boosting machines (GBMs) in terms of RMSE metrics, with a slightly higher MAE. These findings empower insurance companies to enhance risk assessment processes, set suitable premiums, manage reserves, and accurately forecast claim occurrences, contributing to competitive pricing for clients while ensuring profitability. For cross-border trade entities, such as trucking companies and cargo owners, these insights aid in improved risk management and potential cost savings by enabling more reasonable insurance premiums based on accurate predictive claims from insurance companies.

Sequence-developability mapping of affibody and fibronectin paratopes via library-scale variant characterization.

Protein developability is requisite for use in therapeutic, diagnostic, or industrial applications. Many developability assays are low throughput, which limits their utility to the later stages of protein discovery and evolution. Recent approaches enable experimental or computational assessment of many more variants, yet the breadth of applicability across protein families and developability metrics is uncertain. Here, three library-scale assays-on-yeast protease, split green fluorescent protein (GFP), and non-specific binding-were evaluated for their ability to predict two key developability outcomes (thermal stability and recombinant expression) for the small protein scaffolds affibody and fibronectin. The assays' predictive capabilities were assessed via both linear correlation and machine learning models trained on the library-scale assay data. The on-yeast protease assay is highly predictive of thermal stability for both scaffolds, and the split-GFP assay is informative of affibody thermal stability and expression. The library-scale data was used to map sequence-developability landscapes for affibody and fibronectin binding paratopes, which guides future design of variants and libraries.

Development of a Predictive Model to Correlate the Chemical Structure of Amines with Their Oxidative Degradation Rate in a Post-Combustion Amine-Based CO2 Capture Process Using Multiple Linear Regression and Machine Learning Regression Approaches.

In this study, the degradation behavior of 30 different amines was investigated, which were categorized into four distinct groups: alkanolamines, sterically hindered alkanolamines, multialkylamines, and cyclic amines. These experiments were conducted over a period of 14 days at a temperature of 60 °C, with a feed gas comprising 99.9% O2 flowing at a rate of 200 mL/min. The primary objective was to establish a correlation between the chemical structures of these amines and their susceptibilities to degradation. To assess this, the concentration of the amines at various time points was measured to determine their degradation rates. Results showed that secondary amines exhibited degradation rates higher than those of primary and tertiary amines. Amines with cyclic structures demonstrated lower oxidative degradation rates. Longer alkyl chain lengths decreased degradation rates in all amine types because of their electronic and steric hindrance properties. A higher number of hydroxyl groups increased the degradation rate by destabilizing the free radical. An increase in hydroxyl groups in nonsterically hindered amines increased the degradation rate by decreasing free radical stability. In contrast, for sterically hindered amines, an increase in hydroxyl groups decreased the degradation rate because the steric hindrance effect is now more dominant than the electron-withdrawing effect. An increase in the number of amino groups led to higher degradation rates due to the presence of more reactive sites for free radical formation. Amines with tert-alkyl groups exhibited higher degradation rates than those with straight chains. Moreover, branched alkyl groups located between amino and hydroxyl groups significantly increased the degradation rates. Two degradation models, a semiempirical statistical model and a CatBoost machine learning regression model, were developed to predict amine degradation rates based on their chemical structure and relevant properties. To train these models, a data set of 27 different amines was used, while another set of 3 amines was reserved for testing the model's predictive performance. The average absolute deviations (AAD) achieved were, respectively, 22.2 and 0.3%.

Stock Forecasting Based on Linear Regression Model and Nonlinear Machine Learning Regression Model

To enhance the accuracy of stock price prediction for Netflix and provide individuals with a comprehensive understanding of stock trading prices, this study constructs a predictive model by employing three distinct approaches: a linear regression model, a Long Short-term Memory (LSTM) artificial neural network, and a Gated Recursive Unit (GRU) which serves as a component of the LSTM architecture. A prediction scheme is devised, utilizing historical stock data spanning from 2002 to 2022 for Netflix. The primary objective is to forecast the stock price of Netflix for the subsequent 20-day period. To evaluate the efficacy of the three models, a rigorous assessment is conducted employing robust evaluation indices. The outcomes of this analysis will enable a determination of the fitting adequacy of each model, thereby facilitating the identification of the most suitable approach for accurate stock price prediction in the context of Netflix. This research endeavors to contribute to the field of stock market analysis by leveraging advanced predictive modeling techniques for enhanced forecasting accuracy and insightful decision-making.

Comparison of multiple linear regression and machine learning methods in predicting cognitive function in older Chinese type 2 diabetes patients.

The prevalence of type 2 diabetes (T2D) has increased dramatically in recent decades, and there are increasing indications that dementia is related to T2D. Previous attempts to analyze such relationships principally relied on traditional multiple linear regression (MLR). However, recently developed machine learning methods (Mach-L) outperform MLR in capturing non-linear relationships. The present study applied four different Mach-L methods to analyze the relationships between risk factors and cognitive function in older T2D patients, seeking to compare the accuracy between MLR and Mach-L in predicting cognitive function and to rank the importance of risks factors for impaired cognitive function in T2D. We recruited older T2D between 60-95years old without other major comorbidities. Demographic factors and biochemistry data were used as independent variables and cognitive function assessment (CFA) was conducted using the Montreal Cognitive Assessment as an independent variable. In addition to traditional MLR, we applied random forest (RF), stochastic gradient boosting (SGB), Naïve Byer's classifier (NB) and eXtreme gradient boosting (XGBoost). Totally, the test cohort consisted of 197 T2D (98 men and 99 women). Results showed that all ML methods outperformed MLR, with symmetric mean absolute percentage errors for MLR, RF, SGB, NB and XGBoost respectively of 0.61, 0.599, 0.606, 0.599 and 0.2139. Education level, age, frailty score, fasting plasma glucose and body mass index were identified as key factors in descending order of importance. In conclusion, our study demonstrated that RF, SGB, NB and XGBoost are more accurate than MLR for predicting CFA score, and identify education level, age, frailty score, fasting plasma glucose, body fat and body mass index as important risk factors in an older Chinese T2D cohort.

A multiple linear regression-based machine learning model for received signal strength prediction of multiband applications

A multiple linear regression-based machine learning model for received signal strength prediction of multiband applications

ПРОЯВЛЕНИЕ НАРРАТИВНОЙ ЭКОНОМИКИ В США: ВЛИЯНИЕ НА РЫНОЧНУЮ АКТИВНОСТЬ И ВОЛАТИЛЬНОСТЬ

Against the background of growing interest in the role and influence of information noise in economic activity, this study aims to analyze the manifestation of the narrative economy using the example of the United States. The aim of the work is to understand the components of narrative economics and their impact on the market and its participants. This analysis helps to explain which narratives and in which cases incline investors to change their strategies in the market. The object of the study is the growth rate of GDP (economic activity), the profitability of the market and its volatility in the United States. The subject of the study is the influence of narratives on the projected GDP growth rate, profitability and volatility of the US market. The research methodology includes the identification of narratives with a narrative structure using thematic modeling and the use of machine learning methods to identify causeandeffect relationships. Linear regression and machine learning methods, in particular, random forest and XGBoost, were chosen as tools for identifying causeandeffect relationships. The study revealed that narratives have a significant impact on projected GDP growth and economic activity, and there is also a link between market volatility and narratives with an emotional context. However, there is no direct relationship between narratives and market returns. The results obtained may be important for understanding the dynamics of market processes and decisionmaking in the investment sector.

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.

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.

Prediction and Interpretation of Total N and Its Key Drivers in Cultivated Tropical Peat using Machine Learning and Game Theory

Currently, there is a growing interest among research communities in the development of statistical learning-based pedotransfer functions/PtFs to predict mineral soil nutrients; however, similar studies in peatlands are relatively rare. Moreover, extracting meaningful information from these ‘black-box’ models is crucial, particularly concerning their algorithmic complexity and the non-linear nature of the soil covariate interrelationships. This study employed the Pulunggono (2022a) dataset and the bootstrapping method, to (1) develop and evaluate seven PtF models, including both general linear models (GLM) and machine learning (ML) regressors for estimating total nitrogen (N) in tropical peat that has been drained and cultivated for oil palm (OP) in Riau, Indonesia and (2) explaining model functioning by incorporating Shapley Additive Explanation (SHAP), a tool derived from coalitional game theory. This study demonstrated the superior predictive performance of ML-based PtFs in estimating total N compared to GLM algorithms. The top-performing algorithms for PtF models were identified as GBM, XGB, and Cubist. The SHAP method revealed that sampling depth and organic C were consistently identified as the most important covariates across all models, irrespective of their algorithmic capabilities. Additionally, ML algorithms identified the total Fe, pH, and bulk density (BD) as significant covariates. Local explanations based on Shapley values indicated that the behavior of PtF-based algorithms diverged from their global explanations. This study emphasized the critical role of ML algorithms and game theory in accurately predicting total N in peatlands subjected to drainage and cultivation for OP and explaining their model behavior in relation to soil biogeochemical processes.

Differentiation of Sahelian aquifers from chemical and isotopic composition using linear statistics and machine learning

ABSTRACT In Sahelian Africa, the characteristics of boreholes are often lost and, when several aquifers are present on the same site, it is difficult to know which one is being tapped or is likely to be contaminated, which hinders good management of the resource. In this study conducted on 153 wells distributed in the four major aquifers of Burkina Faso, the variation in chemical composition within the aquifers is high compared to that between the aquifers. In spite of this, treatment by linear statistical analysis and/or machine learning allows the discrimination of the aquifers with a success rate of about 80%. The introduction of water isotopes as an additional parameter and a dimensional reduction by principal component analysis allowed a discrimination rate of 87.6% to be achieved. The pathway of water from sedimentary to basement aquifers explains some of the confusion.

Generation of fusion and fusion-evaporation reaction cross-sections by two-step machine learning methods

In order to obtain cross-sections of heavy-ion fusion and fusion-evaporation reactions, artificial neural networks, cubist, random forest, support vector regression, extreme gradient boosting, and multiple linear regression machine learning approaches were used separately in this study. The outcomes from these different methods that are obtained from the training carried out with the existing experimental data in the literature were compared. Furthermore, it has been observed that a two-step process yielded better results for determining the heavy-ion reaction cross-sections, after first estimating which approach would be better for which reaction. In this manner, the method for which the cross-section needs to be calculated is determined by the machine learning classification application, and predictions can be made using the machine learning regression application with the determined method. It has been concluded that the obtained results are in harmony with the experimental data and that the methods can be used safely. The obtained results are published on a web page that allows for online calculation of heavy-ion fusion and fusion-evaporation reaction cross-sections.

Higher heating value prediction of high ash gasification-residues: Comparison of white, grey, and black box models

The measurement of the higher heating value (HHV) in high-ash solid waste poses persistent challenges due to the inherent limitations of using an oxygen bomb calorimeter. HHV predictive models based on the components analysis were proved to be useful methods. To improve the accuracy and applicability of HHV models, this work selected 100 actual measured data of high ash gasification residues, and compared three types of predictive models from white, grey, and black box by the following procedures: Modeling, External-validation, and Extending study. In the Modeling and External-validation processes, eight models from linear regression, grey model, and machine learning models, were proposed with R2 > 0.95 and MAPE<8.42 %. Compared to existing research, the models proposed in this study provide appealing accuracies. In the Extending study for applicability evaluation, the UA-based models UOGM (1, 5) (R2 = 0.86 and MAPE = 12.96 %) and RL-Ult (R2 = 0.89 and MAPE = 11.52 %) outperformed the other six models by using the eight-expanding data of different solid wastes that collected from the literature. Based on these results, the above two models were applied to predict the HHV of remaining 53 gasifier residues, showing reliable predicting results. This work shows that the developed models have high accuracy for the HHV prediction of high-ash solid wastes.

Multiple Stock Prediction Based on Linear and Non-linear Machine Learning Regression Methods

In contemporary times, the pressing issue of global environmental pollution has prompted the exploration of alternative energy sources by various industries, aiming to mitigate the adverse environmental impacts caused by traditional energy production. Correspondingly, investors in the financial market have increasingly redirected their capital towards the new energy sector. Within this context, the present research endeavors to employ machine learning techniques for the prediction of Tesla's stock price. This study leverages multiple linear regression, polynomial regression, and lag models to construct models based on the datasets of TSLA, MPC, and UNG stock prices spanning the period of 2019-2020. By discerning potential patterns among these variables, the objective is to anticipate the future trajectory of TSLA stock price. According to machine learning methods, Tesla's stock price can be predicted, and the daily price of Tesla is influenced by the opening price, high price, low price and trading volume of the stock on that day. In addition, the share prices of energy companies related to Tesla also have an impact on Tesla's share price on that day. Specifically, Tesla's stock price is influenced by Natural Gas Company (UNG), which has an opposite relationship. Although common sense economics says that the crude oil market will be closely related to the new energy market. However, the results of this study demonstrated that Tesla's stock price is less influenced by Crude Oil Company (MPC).

Snapshots of daily life: Situations investigated through the lens of smartphone sensing.

Daily life unfolds in a sequence of situational contexts, which are pivotal for explaining people's thoughts, feelings, and behaviors. While situational data were previously difficult to collect, the ubiquity of smartphones now opens up new opportunities for assessing situations in situ, that is, while they occur. Seizing this opportunity, the present study demonstrates how smartphones can help establish associations between the psychological perception and physical reality of situations. We employed an intensive longitudinal sampling design and investigated 9,790 situational snapshots experienced by 455 participants for 14 consecutive days. These snapshots combined self-reported situation characteristics from experience samplings with their corresponding objective cues obtained via smartphone sensing. More precisely, we extracted a total of 1,356 granular cues from different sensing modalities to account for the complexity of real-world situations. We applied linear and nonlinear machine learning algorithms to examine how well these cues predicted the perceived characteristics in terms of the Situational Eight Duty, Intellect, Adversity, Mating, pOsitivity, Negativity, Deception, Sociality (DIAMONDS), finding significant out-of-sample predictions for the five dimensions reflecting the situations' Duty, Intellect, Mating, pOsitivity, and Sociality. In a series of follow-up analyses, we further explored the data patterns captured by our models, revealing, for example, that those cues related to time and location were particularly informative of the respective situation characteristics. We conclude by interpreting the mapping between cues and characteristics in real-world situations and discussing how smartphone-based situational snapshots may push the boundaries of psychological research on situations. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Biomass gasification technology for a distinct renewable electricity/heat production scheme; Developed regression models based on machine learning algorithms for power/hot water/efficiency/emission data analysis

Rice straw is a promising feedstock for gasification, and using it in combined heat and power systems allows us to generate renewable electricity and heat while reducing reliance on fossil fuels and combatting climate change. As the demand for clean energy grows, machine learning will become even more crucial for managing our energy systems. A combined heat and power system was integrated based on rice straw gasification and the applications of linear and quadratic regression machine learning algorithms were evaluated using the residual analysis and the analysis of variance. The results show that quadratic models outperform linear models in all cases. Specifically, for efficiency modeling, the quadratic model scores 98.2 %, surpassing the linear model's 89.6 %. In emission modeling, the quadratic model outperforms the linear model with an R-sq score of 97 % versus 88.1 %. The results show that equivalence ratio of gasifier has the greatest impact, accounting for 65 % of the observed variation on emission linear model. In the quadratic emission model, linear factors contribute 58 % and quadratic factors contribute 42 %. Among the linear factors, pressure ratio has a significant influence with a contribution rate of 89 %. In terms of quadratic factors, pressure ratio has the highest influence, with a remarkable contribution rate of 92 %. This research emphasizes the significance of enhancing the precision of regression machine learning algorithms by conducting residual analysis and variance analysis.

Association of genetic and sulcal traits with executive function in congenital heart disease.

Persons with congenital heart disease (CHD) are at increased risk of neurodevelopmental disabilities, including impairments to executive function. Sulcal pattern features correlate with executive function in adolescents with single-ventricle heart disease and tetralogy of Fallot. However, the interaction of sulcal pattern features with genetic and participant factors in predicting executive dysfunction is unknown. We studied sulcal pattern features, participant factors, and genetic risk for executive function impairment in a cohort with multiple CHD types using stepwise linear regression and machine learning. Genetic factors, including predicted damaging de novo or rare inherited variants in neurodevelopmental disabilitiesrisk genes, apolipoprotein E genotype, and principal components of sulcal pattern features were associated with executive function measures after adjusting for age at testing, sex, mother's education, and biventricular versus single-ventricle CHD in a linear regression model. Using regression trees and bootstrap validation, younger participant age and larger alterations in sulcal pattern features were consistently identified as important predictors of decreased cognitive flexibility with left hemisphere graph topology often selected as the most important predictor. Inclusion of both sulcal pattern and genetic factors improved model fit compared to either alone. We conclude that sulcal measures remain important predictors of cognitive flexibility, and the model predicting executive outcomes is improved by inclusion of potential genetic sources of neurodevelopmental risk. If confirmed, measures of sulcal patterning may serve as early imaging biomarkers to identify those at heightened risk for future neurodevelopmental disabilities.