Software defects prediction by metaheuristics tuned extreme gradient boosting and analysis based on Shapley Additive Explanations

Applied Soft Computing, Journal Year: 2023, Volume and Issue: 146, P. 110659 - 110659

Published: July 29, 2023

Language: Английский

---

Machine Learning to Develop Credit Card Customer Churn Prediction

Journal of theoretical and applied electronic commerce research, Journal Year: 2022, Volume and Issue: 17(4), P. 1529 - 1542

Published: Nov. 16, 2022

The credit card customer churn rate is the percentage of a bank’s customers that stop using services. Hence, developing prediction model to predict expected status for will generate an early alert banks change service or offer them new This paper aims develop by feature-selection method and five machine learning models. To select independent variables, three models were used, including selection all two-step clustering k-nearest neighbor, feature selection. In addition, selected, Bayesian network, C5 tree, chi-square automatic interaction detection (CHAID) classification regression (CR) neural network. analysis showed could model. results tree performed best in comparison with developed indicated top variables needed development total transaction count, revolving balance on card, count. Finally, revealed merging multi-categorical into one variable improved performance

Language: Английский

---

Addressing Internet of Things security by enhanced sine cosine metaheuristics tuned hybrid machine learning model and results interpretation based on SHAP approach

PeerJ Computer Science, Journal Year: 2023, Volume and Issue: 9, P. e1405 - e1405

Published: June 30, 2023

An ever increasing number of electronic devices integrated into the Internet Things (IoT) generates vast amounts data, which gets transported via network and stored for further analysis. However, besides undisputed advantages this technology, it also brings risks unauthorized access data compromise, situations where machine learning (ML) artificial intelligence (AI) can help with detection potential threats, intrusions automation diagnostic process. The effectiveness applied algorithms largely depends on previously performed optimization, i.e., predetermined values hyperparameters training conducted to achieve desired result. Therefore, address very important issue IoT security, article proposes an AI framework based simple convolutional neural (CNN) extreme (ELM) tuned by modified sine cosine algorithm (SCA). Not withstanding that many methods addressing security issues have been developed, there is always a possibility improvements proposed research tried fill in gap. introduced was evaluated two ToN intrusion datasets, consist traffic generated Windows 7 10 environments. analysis results suggests model achieved superior level classification performance observed datasets. Additionally, conducting rigid statistical tests, best derived interpreted SHapley Additive exPlanations (SHAP) findings be used experts enhance systems.

Language: Английский

---

Improving audit opinion prediction accuracy using metaheuristics-tuned XGBoost algorithm with interpretable results through SHAP value analysis

Applied Soft Computing, Journal Year: 2023, Volume and Issue: 149, P. 110955 - 110955

Published: Oct. 21, 2023

Language: Английский

---

Marine Vessel Classification and Multivariate Trajectories Forecasting Using Metaheuristics-Optimized eXtreme Gradient Boosting and Recurrent Neural Networks

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(16), P. 9181 - 9181

Published: Aug. 11, 2023

Maritime vessels provide a wealth of data concerning location, trajectories, and speed. However, while these are meticulously monitored logged to maintain course, they can also meta information. This work explored the potential data-driven techniques applied artificial intelligence (AI) tackle two challenges. First, vessel classification was through use extreme gradient boosting (XGboost). Second, trajectory time series forecasting tackled long-short-term memory (LSTM) networks. Finally, due strong dependence AI model performance on proper hyperparameter selection, boosted version well-known particle swarm optimization (PSO) algorithm introduced specifically for tuning hyperparameters models used in this study. The methodology real-world automatic identification system (AIS) both marine forecasting. Boosted PSO (BPSO) compared contemporary optimizers showed promising outcomes. XGBoost tuned using attained an overall accuracy 99.72% problem, LSTM mean square error (MSE) 0.000098 prediction challenge. A rigid statistical analysis performed validate outcomes, explainable principles were determined best-performing models, gain better understanding feature impacts decisions.

Language: Английский

---

Applying Recurrent Neural Networks for Anomaly Detection in Electrocardiogram Sensor Data

Sensors, Journal Year: 2023, Volume and Issue: 23(24), P. 9878 - 9878

Published: Dec. 17, 2023

Monitoring heart electrical activity is an effective way of detecting existing and developing conditions. This usually performed as a non-invasive test using network up to 12 sensors (electrodes) on the chest limbs create electrocardiogram (ECG). By visually observing these readings, experienced professionals can make accurate diagnoses and, if needed, request further testing. However, training experience needed are significant. work explores potential recurrent neural networks for anomaly detection in ECG readings. Furthermore, attain best possible performance networks, parameters, architectures optimized modified version well-established particle swarm optimization algorithm. The models compared created by other contemporary optimizers, results show significant real-world applications. Further analyses carried out best-performing determine feature importance.

Language: Английский

---

A Machine Learning Method with Hybrid Feature Selection for Improved Credit Card Fraud Detection

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(12), P. 7254 - 7254

Published: June 18, 2023

With the rapid developments in electronic commerce and digital payment technologies, credit card transactions have increased significantly. Machine learning (ML) has been vital analyzing customer data to detect prevent fraud. However, presence of redundant irrelevant features most real-world degrades performance ML classifiers. This study proposes a hybrid feature-selection technique consisting filter wrapper steps ensure that only relevant are used for machine learning. The proposed method uses information gain (IG) rank features, top-ranked fed genetic algorithm (GA) wrapper, which extreme (ELM) as algorithm. Meanwhile, GA is optimized imbalanced classification using geometric mean (G-mean) fitness function instead conventional accuracy metric. approach achieved sensitivity specificity 0.997 0.994, respectively, outperforming other baseline techniques methods recent literature.

Language: Английский

---

Machine learning tuning by diversity oriented firefly metaheuristics for Industry 4.0

Expert Systems, Journal Year: 2023, Volume and Issue: 41(2)

Published: March 30, 2023

Abstract The progress of Industrial Revolution 4.0 has been supported by recent advances in several domains, and one the main contributors is Internet Things. Smart factories healthcare have both benefited terms leveraged quality service productivity rate. However, there always a trade‐off some largest concerns include security, intrusion, failure detection, due to high dependence on Things devices. To overcome these other challenges, artificial intelligence, especially machine learning algorithms, are employed for fault prediction, intrusion computer‐aided diagnostics, so forth. efficiency models heavily depend feature selection, predetermined values hyper‐parameters training deliver desired result. This paper proposes swarm intelligence‐based approach tune models. A novel version firefly algorithm, that overcomes known deficiencies original method employing diversification‐based mechanism, proposed applied selection hyper‐parameter optimization two models—XGBoost extreme machine. tested four real‐world Industry data sets, namely distributed transformer monitoring, elderly fall BoT‐IoT, UNSW‐NB 15. Achieved results compared eight cutting‐edge metaheuristics, implemented under same conditions. experimental outcomes strongly indicate significantly outperformed all competitor metaheuristics convergence speed results' measured with standard metrics—accuracy, precision, recall, f1‐score.

Language: Английский

---

Addressing feature selection and extreme learning machine tuning by diversity-oriented social network search: an application for phishing websites detection

Complex & Intelligent Systems, Journal Year: 2023, Volume and Issue: 9(6), P. 7269 - 7304

Published: June 28, 2023

Abstract Feature selection and hyper-parameters optimization (tuning) are two of the most important challenging tasks in machine learning. To achieve satisfying performance, every learning model has to be adjusted for a specific problem, as efficient universal approach does not exist. In addition, data sets contain irrelevant redundant features that can even have negative influence on model’s performance. Machine applied almost everywhere; however, due high risks involved with growing number malicious, phishing websites world wide web, feature tuning this research addressed particular problem. Notwithstanding many metaheuristics been devised both challenges, there is still much space improvements. Therefore, exhibited manuscript tries improve website detection by extreme utilizes relevant subset features. accomplish goal, novel diversity-oriented social network search algorithm developed incorporated into two-level cooperative framework. The proposed compared six other cutting-edge algorithms, were also implemented framework tested under same experimental conditions. All employed level 1 perform task. best-obtained then used input 2, where all algorithms machine. Tuning referring neurons hidden layers weights biases initialization. For evaluation purposes, three different sizes classes, retrieved from UCI Kaggle repositories, methods terms classification error, separately 2 over several independent runs, detailed metrics final outcomes (output layer 2), including precision, recall, f1 score, receiver operating characteristics precision–recall area curves. Furthermore, an additional experiment conducted, only used, establish performance features, which represents large-scale NP-hard global challenge. Finally, according results statistical tests, findings suggest average obtains better achievements than competitors challenges sets. SHapley Additive exPlanations analysis best-performing was determine influential

Language: Английский

---

Improving performance of extreme learning machine for classification challenges by modified firefly algorithm and validation on medical benchmark datasets

Multimedia Tools and Applications, Journal Year: 2024, Volume and Issue: 83(31), P. 76035 - 76075

Published: Feb. 21, 2024

Language: Английский

---