Effectiveness Of Machine Learning Models Research Articles

Designing graph neural networks training data with limited samples and small network sizes

Machine learning is a data-driven domain, which means a learning model's performance depends on the availability of large volumes of data to train it. However, by improving data quality, we can train effective machine learning models with little data. This paper demonstrates this possibility by proposing a methodology to generate high-quality data in the networking domain. We designed a dataset to train a given Graph Neural Network (GNN) that not only contains a small number of samples, but whose samples also feature network graphs of a reduced size (10-node networks). Our evaluations indicate that the dataset generated by the proposed pipeline can train a GNN model that scales well to larger networks of 50 to 300 nodes. The trained model compares favorably to the baseline, achieving a mean absolute percentage error of 5-6%, while being significantly smaller at 90 samples total (vs. thousands of samples for the baseline).

Predicting primary and middle-school students’ preferences for online learning with machine learning

Background: The COVID-19 pandemic has brought attention to student psychological wellness. Because of isolation, lack of socialisation and intellectual and physical development from excessive media use, primary and secondary school students are at high risk for health problems.Aim: This study aimed to identify the most effective machine learning model for predicting the offline and online instructional strategies students would choose during a pandemic.Setting: The study was carried out at a number of primary and middle schools in Hyderabad, India.Methods: We evaluated the data using machine learning methods such as logistic regression, K-nearest neighbour (KNN), decision trees, bagging and boosting using the Python programming language.Results: In this study, 414 instances were collected from different schools. Exploratory data analysis showed that few students chose online courses. According to the research, very few students choose online classes, and the majority of students favoured offline classes over online because of physical and mental health difficulties; online education effects include a lack of social and peer relationships that affects young children psychologically, and they may not be disciplined enough to resist internet diversions. Smartphones, laptops, etc., affect their vision, causing headaches and impaired eyesight.Conclusion: The KNN was the most accurate machine learning algorithm, with 92.13% accuracy to fits the data to identify the preferences of online education.Contribution: This article examined the perspectives of primary and middle-school children on online education. Most students in this survey also reported experiencing mental or physical health issues that made online education difficult for them. Machine learning algorithms were applied to identify the most effective model for predicting students’ online and offline study preferences. This machine learning method will help schools improve their course delivery methods, allowing students to continue their studies without interruption.

The role of higher moments in predicting China's oil futures volatility: Evidence from machine learning models

This paper expands the emerging literature on volatility forecasting for China's oil market by exploring the predictive ability of higher-order moments (skewness, kurtosis, hyperskewness, and hyperkurtosis) based on high-frequency data. Our investigation is originally based on the heterogeneous autoregressive (HAR) framework, but considering the possible multicollinearity and nonlinearity, it is extended to various machine learning (ML) models and combination forecasting models. The results reveal that higher-order moments, including the two highest moments, always significantly improve predictive performance for the COVID-19 crisis. We further examine the interpretability of ML models and each factor's contribution to the prediction, finding that odd and even moments contain short- and long-term prediction information, respectively. This paper also highlights the effectiveness of ML models for capturing trends in oil futures volatility with higher-order moments and the satisfactory performance of combination forecasting models. Finally, we investigate the predictability of asymmetric risk patterns and obtain identical results. Our study has important implications for financial risk management, asset pricing, and portfolio allocation.

An efficient convergence-boosted salp swarm optimizer-based artificial neural network for the development of software fault prediction models

Machine learning (ML) approaches were employed to tackle the software fault prediction (SFP) issue due to their consistent and rigorous performance. Multilayer perceptron (MLP) neural networks are one of the most effective ML models for prediction problems. Unfortunately, MLP suffers from chronic shortcomings related to the gradient-descent learning mechanism that can easily get stuck in local minima, leading to inappropriate control parameters. For SFP tackling, MLP boosted with an improved version of Salp Swarm Optimizer (SSA), a metaheuristic swarm intelligence technique, is presented in this research. The MLP learning approach is updated with SSA to remedy these flaws. The key benefit of such an algorithm is its ability to avoid local minima through convergence behavior. Two improvements were developed in the SSA optimization loop to link SSA functionality with MLP. The first improvement is elitism (SSA-elitism), while the second is MSSA, which stands for search improvement. The performance of proposed SSA versions is evaluated using 18 benchmark SFP datasets using ROC, sensitivity, specificity, and accuracy performance metrics. Several evaluations and validations were performed by contrasting the results of the developed versions with those of the conventional MLP, SSA, and 10 state-of-the-art approaches. The evaluations and validation findings prove that the proposed versions have superior capabilities to efficiently optimize MLP parameters, which raises the quality of their predictability.

Classification and Regression Using Automatic Machine Learning (AutoML) – Open Source Code for Quick Adaptation and Comparison

This paper presents a comprehensive exploration of automatic machine learning (AutoML) tools in the context of classification and regression tasks. The focus lies on understanding and illustrating the potential of these tools to accelerate and optimize the process of machine learning, thereby making it more accessible to non-experts. Specifically, we delve into multiple popular open-source AutoML tools and provide illustrative examples of their application. We first discuss the fundamental principles of AutoML, including its key features such as automated data preprocessing, feature engineering, model selection, hyperparameter tuning, and model validation. We subsequently venture into the hands-on application of these tools, demonstrating the implementation of classification and regression tasks using multiple open-source AutoML tools. We provide open-source code samples for two data scenarios for classification and regression, designed to assist readers in quickly adapting AutoML tools for their own projects and in comparing the performance of different tools. We believe that this contribution will aid both practitioners and researchers in harnessing the power of AutoML for efficient and effective machine learning model development.

A machine learning model for fast approximation of free-surface Green's function and its application

In potential flow theory, the accurate and efficient calculation of free-surface Green's functions is essential for solving hydrodynamic issues. Given the impressive performance of machine learning methods in nonlinear function fitting, the present study utilizes an effective machine learning model called StripeGF for numerical approximation. In this model, equidistant horizontal datum lines are arranged in the computational domain away from the singularity, and Green's function and its derivatives on each line are fitted by a multi-layer perceptron (MLP) with a single input. Based on the first-order ordinary differential equation (ODE) that they satisfy, the fourth-order Runge-Kutta method is used to solve the Green's function and its derivatives between adjacent lines. In the domain nearing the singularity, a double-input MLP is applied. Use the Romberg quadrature to create a double-precision data set for training and validation, the numerical results demonstrate that StripeGF outperforms all 4 comparison methods in terms of efficiency and has accuracy of at least 4 digits in more than 99.9% of all zones. The boundary element program improved by StripeGF is verified in the hydrodynamic calculation of S175, showing good accuracy and reliability.

Machine learning for accurate methane concentration predictions: short-term training, long-term results

Abstract Although methane emissions from Alberta’s oil and gas sector have decreased in recent years, monitoring these emissions using Continuous Emission Monitoring Systems (CEMS) can be costly. Predictive Emissions Monitoring Systems (PEMS), powered by machine learning, offer an alternative to or can supplement CEMS. However, effective machine learning models for methane emissions prediction rely heavily on the amount of training data. To address this, we compare the prediction performance of different neural network models, including Long Short-Term Memory (LSTM), Stacked LSTM, Gated Recurrent Unit (GRU), and Bidirectional LSTM (BiLSTM), using varying time intervals for training of methane concentration data from Alberta airshed stations. The results showed that the GRU model performed better with shorter datasets, whereas the LSTM and Stacked LSTM models outperformed the GRU and BiLSTM models when trained with more historical data. However, the study found that more training data did not necessarily result in significantly better prediction models.

Modeling and Mapping of Flood Susceptibility at Que Son District, Quang Nam Province, Vietnam using CatBoost

In this research, the main objective is to model and map flood susceptibility in Que Son district, Quang Nam province, Vietnam using one of the effective machine learning model namely CatBoost. With this purpose, a total of 96 flood and non-flood locations and a set of 10 conditioning factors were collected to construct the geospatial database. Thereafter, Shap feature importance method was used to validate and select the most important conditioning factors used for modeling of flood susceptibility, and the results showed that only 8 conditioning factors including aspect, slope, curvature, elevation, land cover, rainfall, distance to rivers, and Topographic Wetness Index (TWI) were selected for final modelling of flood susceptibility at the study area. Validation of the model was also done using various statistical indexes including area under the ROC curve (AUC). Validation results showed that the performance of CatBoost model (AUC = 0.96 for training and AUC = 0.94 for testing) is good for prediction of flood susceptibility of the study area. Thus, it can be concluded that CatBoost is valuable tool for flood susceptibility modeling which can be used to assess flood susceptibility in other flood prone areas of the world. In addition, flood susceptibility map generated from CatBoost model in this study might be helpful in development of better flood mitigation strategies at the study area.

PARSIMONIOUS MACHINE LEARNING MODELS IN REQUIREMENTS ELICITATION TECHNIQUES SELECTION

The subject of research in the article is machine learning algorithms used for requirement elicitation technique selection. The goal of the work is to build effective parsimonious machine learning models to predict the using particular elicitation techniques in IT projects that allow using as few predictor variables as possible without a significant deterioration in the prediction quality. The following tasks are solved in the article: design an algorithm to build parsimonious machine learning candidate models for requirement elicitation technique selection based on gathered information on practitioners' experience, assess parsimonious machine learning model accuracy, and design an algorithm for the best candidate model selection. The following methods are used: algorithm theory, statistics theory, sampling techniques, data modeling theory, and science experiments. The following results were obtained: 1) parsimonious machine learning candidate models were built for the requirement elicitation technique selection. They included less number of features that helps in the future to avoid overfitting problems associated with the best-fit models; 2) according to the proposed algorithm for best candidate selection – a single parsimonious model with satisfied performance was chosen. Conclusion: An algorithm is proposed to build parsimonious candidate models for requirement elicitation technique selection that avoids the overfitting problem. The algorithm for the best candidate model selection identifies when a parsimonious model's performance is degraded and decides on the suitable model's selection. Both proposed algorithms were successfully tested with four datasets and can be proposed for their extensions to others.

Hybrid ECBO–ANN Algorithm for Shear Strength of Partially Grouted Masonry Walls

In recent years, artificial neural network (ANN) has become one of the popular and effective machine learning models, having a unique ability to handle very complex problems and the potential to predict accurate results without a defined algorithmic solution. However, the ANN structure and parameters are usually chosen by experience.The behavior of Partially Grouted (PG) masonry shear walls is complex due to the inherent anisotropic properties of the masonry materials and the nonlinear interactions between mortar, blocks, grouted cells, non-grouted cells, and reinforcing steel.In this study, the aim is to develop an artificial neural network model by combining the ECBO meta-heuristic algorithm with the artificial neural network structure to optimize the feed forward propagation network parameters for analyzing the shear strength of PG walls.A total of 255 test data on PG collected from the available literature were used to generate training and test data sets. Various validation criteria such as mean square error, root mean square error and correlation coefficient (R) are used to validate the models.In this study, the optimal number of neurons used in the hidden layer and also the optimal number of CBs required in the ECBO algorithm were obtained. The mathematical formulation of the optimized neural network model with the combination of meta-heuristic algorithm is also presented.

Mathematical Modelling of the Drying Characteristics of Indian Spinach Using Machine Learning Approach

This study employs machine learning techniques to investigate the drying characteristics of Indian spinach leaves. Drying experiments were conducted at varying air temperatures (50°C, 60°C, 70°C) and velocities (1.2m/s and 1.4m/s) while monitoring drying time, moisture content, and important parameters. Artificial neural networks and support vector machine were used to predict drying rate and final moisture content based on conditions. Higher temperatures and air velocities lead to shorter drying times and lower moisture contents, with higher air velocities enhancing moisture diffusion and reducing energy intensity. Nine drying models were fitted, with the Page and Hii, et al. models showing strong performance. The MultiLayer Perceptron (MLP) 3-8-1 configuration achieves high accuracy, and Support Vector Machine (SVM) models with the Radial Basis Function (RBF) kernel perform well in predicting moisture ratio. These findings offer insights into Indian spinach drying behavior and highlight the effectiveness of machine learning models. The obtained values aid in optimizing drying conditions, and the robust performance of MLP and SVM models holds potential for applications beyond drying, including diabetic retinopathy diagnosis.

Could You Understand Me? The Relationship among Method Complexity, Preprocessing Complexity, Interpretability, and Accuracy

The need to train experts who will be able to apply machine learning methods for knowledge discovery is increasing. Building an effective machine learning model requires understanding the principle of operation of the individual methods and their requirements in terms of data pre-preparation, and it is also important to be able to interpret the acquired knowledge. This article presents an experiment comparing the opinion of the 42 students of the course called Introduction to Machine Learning on the complexity of the method, preprocessing, and interpretability of symbolic, subsymbolic and statistical methods with the correctness of individual methods expressed on the classification task. The methodology of the implemented experiment consists of the application of various techniques in order to search for optimal models, the accuracy of which is subsequently compared with the results of a knowledge test on machine learning methods and students’ opinions on their complexity. Based on the performed non-parametric and parametric statistic tests, the null hypothesis, which claims that there is no statistically significant difference in the evaluation of individual methods in terms of their complexity/demandingness, the complexity of data preprocessing, the comprehensibility of the acquired knowledge and the correctness of the classification, is rejected.

A federated learning framework for pneumonia image detection using distributed data

Pneumonia is one of the serious diseases affecting the lungs. Yearly, over four million people die on average. Therefore, it is essential to have an effective system for early diagnoses. State-of-the-art computer-aided Machine Learning (ML) techniques have been used for pneumonia detection. However, pneumonia X-ray images are visually heterogeneous and complex in pattern recognition. Therefore, a vast amount of dataset is required for effective ML model training. The larger data volume can be collected using the real-time dataset from hospitals and medical institutions. However, due to General Data Protection Regulation (GDPR) and the Data Protection Act (DPA), data sharing is not allowed by the third party. This study is inspired by using real-time datasets in a privacy-preserving fashion while using the framework of federated learning (FL). We have performed experiments using state-of-the-art ML models for medical image classification, including pre-trained Convolutional Neural Network (CNN) models of Alexnet, DenseNet, Residual Neural Network-50 (ResNet50), Inception, and Visual Geometry Group-19 (VGG19). The experiments are performed individually on the models and the FL framework. We compared the results using the evaluation metrics and Area Under the Curve (AUC). The preliminary results show the ResNet-50 stands out in performance on the testing dataset producing an accuracy of 93% significantly.

최신의 하이퍼파라미터 최적화 프레임워크들의 상대적 성능 평가

Machine learning (ML) has proven to be highly effective in solving complex problems in various domains, thanks to its ability to identify specific data tasks, perform feature engineering, and learn quickly. However, designing and training ML models is a complicated task and requires optimization. The effectiveness of ML models is highly dependent on the selection of hyperparameters that determines their performance. Hyperparameter optimization (HPO) is a systematic search process to find the optimal combinations of hyperparameters to achieve robust performance. Traditional HPO methods such as grid and random search take a lot of computing time when used in large-scale applications. Recently, various automated search strategies, such as Bayesian optimization (BO) and evolutionary algorithms, have been developed to significantly reduce the computing time. In this paper, we use state-of-the-art HPO frameworks, namely BO, Optuna, HyperOpt, and Keras Tuner, for optimizing the ML and deep learning models for the classification tasks and evaluate their comparative performance using two different sets of experiments. The first one uses different ML classifiers to solve the optimal parameter selection problem with HPO. The second one attempts to optimize the convolutional neural network (CNN) architecture using HPO frameworks to improve its performance in the image classification task. We use four publicly available real-world datasets including one image dataset. The experimental results show that HyperOpt - TPE outperforms the other HPO frameworks for the ML classifiers and achieves up to 94.12% of accuracy with 30 minutes for performing the optimization. Similarly, for the CNN model, HyperOpt-TPE outperforms the other HPO frameworks by improving 34% of the classification accuracy, while taking 2 hours and 24 minutes of computing time.

Machine learning algorithm to predict the in-hospital mortality in critically ill patients with chronic kidney disease

Background This study aimed to establish and validate a machine learning (ML) model for predicting in-hospital mortality in critically ill patients with chronic kidney disease (CKD). Methods This study collected data on CKD patients from 2008 to 2019 using the Medical Information Mart for Intensive Care IV. Six ML approaches were used to build the model. Accuracy and area under the curve (AUC) were used to choose the best model. In addition, the best model was interpreted using SHapley Additive exPlanations (SHAP) values. Results There were 8527 CKD patients eligible for participation; the median age was 75.1 (interquartile range: 65.0–83.5) years, and 61.7% (5259/8527) were male. We developed six ML models with clinical variables as input factors. Among the six models developed, the eXtreme Gradient Boosting (XGBoost) model had the highest AUC, at 0.860. According to the SHAP values, the sequential organ failure assessment score, urine output, respiratory rate, and simplified acute physiology score II were the four most influential variables in the XGBoost model. Conclusions In conclusion, we successfully developed and validated ML models for predicting mortality in critically ill patients with CKD. Among all ML models, the XGBoost model is the most effective ML model that can help clinicians accurately manage and implement early interventions, which may reduce mortality in critically ill CKD patients with a high risk of death.

A holistic and proactive approach to forecasting cyber threats

Traditionally, cyber-attack detection relies on reactive, assistive techniques, where pattern-matching algorithms help human experts to scan system logs and network traffic for known virus or malware signatures. Recent research has introduced effective Machine Learning (ML) models for cyber-attack detection, promising to automate the task of detecting, tracking and blocking malware and intruders. Much less effort has been devoted to cyber-attack prediction, especially beyond the short-term time scale of hours and days. Approaches that can forecast attacks likely to happen in the longer term are desirable, as this gives defenders more time to develop and share defensive actions and tools. Today, long-term predictions of attack waves are mostly based on the subjective perceptiveness of experienced human experts, which can be impaired by the scarcity of cyber-security expertise. This paper introduces a novel ML-based approach that leverages unstructured big data and logs to forecast the trend of cyber-attacks at a large scale, years in advance. To this end, we put forward a framework that utilises a monthly dataset of major cyber incidents in 36 countries over the past 11 years, with new features extracted from three major categories of big data sources, namely the scientific research literature, news, blogs, and tweets. Our framework not only identifies future attack trends in an automated fashion, but also generates a threat cycle that drills down into five key phases that constitute the life cycle of all 42 known cyber threats.

Exploiting similarity-induced redundancies in correlation topology for channel pruning in deep convolutional neural networks

The paper discusses the high computational costs associated with convolutional neural networks (CNNs) in real-world applications due to their complex structure, primarily in hidden layers. To overcome this issue, the paper proposes a novel channel pruning technique that leverages the correlation topology of feature maps generated by each CNNs layer to construct a network with fewer nodes, reducing computational costs significantly. Redundant channels exhibit a high degree of topological similarity and tend to increase as the number of network layers rises. Removing the channel corresponding to highly correlated feature maps allows retrieval of the ‘base’ set of characteristics needed by subsequent layers. The proposed channel pruning technique provides a promising approach to reducing the computational costs of deep convolutional neural networks while maintaining high performance levels. By designing a network structure optimized for specific input data types, the method results in more efficient and effective machine learning models. The pruning operation requires fine-tuning to optimize network performance, and experiments using X-ray, chest CT, and MNIST images show that the pruned network can eliminate approximately 80% of redundant channels with minimal performance deterioration (maintaining original CNNs performance at 99.2%).

Exploring the Role of Machine Learning in Contact Tracing for Public Health: Benefits, Challenges, and Ethical Considerations

This article discusses the role of machine learning in contact tracing for public health, particularly in controlling the spread of infectious diseases like COVID-19. The traditional methods of contact tracing have proven to be insufficient in dealing with the scale and complexity of the pandemic, leading to the exploration of new technologies such as machine learning. The article reviews various machine learning models and techniques that have been developed for contact tracing and highlights the potential benefits of using machine learning, including improved accuracy, efficiency, and personalized risk assessments. One of the major challenges in contact tracing using machine learning is data collection and integration. The article discusses the importance of data quality and integration in developing effective machine learning models. It also highlights the need for privacy and security protocols to protect sensitive data and ensure ethical use of machine learning in contact tracing. The article also discusses various evaluation metrics and techniques that can be used to assess the performance of machine learning models in contact tracing. The ethical considerations and challenges of using machine learning in contact tracing are also discussed in detail, highlighting the importance of developing ethical frameworks that can guide the use of machine learning in contact tracing, ensuring that it is both effective and socially responsible. The article provides case studies of the use of machine learning in contact tracing, including lessons learned and best practices. The article discusses future directions and opportunities for the use of machine learning in contact tracing, highlighting the need for ongoing research and development in this area to improve the accuracy, efficiency, and accessibility of machine learning models for contact tracing. This article provides a comprehensive overview of the potential benefits of using machine learning in contact tracing and the challenges and ethical considerations that must be addressed.

All-natural phyllosilicate-polysaccharide triboelectric sensor for machine learning-assisted human motion prediction

The rapid development of smart and carbon-neutral cities motivates the potential of natural materials for triboelectric electronics. However, the relatively deficient charge density makes it challenging to achieve high Maxwell’s displacement current. Here, we propose a methodology for improving the triboelectricity of marine polysaccharide by incorporating charged phyllosilicate nanosheets. As a proof-of-concept, a flexible, flame-retardant, and eco-friendly triboelectric sensor is developed based on all-natural composite paper from alginate fibers and vermiculite nanosheets. The interlaced fibers and nanosheets not only enable superior electrical output but also give rise to wear resistance and mechanical stability. The fabricated triboelectric sensor successfully monitors slight motion signals from various joints of human body. Moreover, an effective machine-learning model is developed for human motion identification and prediction with accuracy of 96.2% and 99.8%, respectively. This work offers a promising strategy for improving the triboelectricity of organo-substrates and enables implementation of self-powered and intelligent platform for emerging applications.

Efficient training of two ANNs using four meta-heuristic algorithms for predicting the FRP strength

In recent years, artificial neural network (ANN) is one of the popular and effective machine learning models that can be used to accurately predict fiber reinforced polymer (FRP) strength. However, the ANN structure and parameters are usually are chosen by experience. In this study, the aim is to use the combination of meta-heuristic algorithm with two different types of artificial neural network structure to optimize the parameters of the feed forward backpropagation and radial basis function networks. In this paper, Particle swarm optimization (PSO), Genetic algorithm (GA), Colliding bodies optimization (CBO), Enhanced colliding bodies optimization (ECBO) algorithms are used to combine with ANNs.A total of 223 test data on Carbon FRP (CFRP) collected from the available literature were used to generate training and test data sets. Various validation criteria such as mean square error, root mean square error and correlation coefficient (R) are used to validate the models. These models consider the effects of concrete compressive strength, concrete sample diameter, concrete sample length, fiber elastic modulus, fiber thickness, fiber strength on the ultimate strength of FRP-concrete.It is shown that ECBO algorithm provide better results and higher accuracy. Also, the comparison of the results of using two neural networks, feed forward backpropagation (FFB) and radial basis function (RBF), indicates that the lower error percentage and more accurate results are related to the FFB neural networks combined with ECBO.