The rapid increase in smart devices and the Internet of Things, which allows sharing and receiving data with other devices and systems over the Internet, increases cyber security threats. These devices are becoming prime targets for cyber attacks. Furthermore, detecting these attacks by using traditional techniques is becoming a challenge. We switched to Deep Learning and Machine Learning techniques to address these challenges. In short words, we used AI-driven approaches to recover from these challenges.
Introduction:
The Internet of Things(IoT), the persistent connection of tangible entities with the Internet constitutes new cyber security risks and demands monitoring and appraisal. Anticipating the types of attacks and preventing data from unforeseen attacks are the utmost requirements of cybersecurity. Obstructing data from being attacked by malware is a crucial step toward cybersecurity. Anyhow, the traditional ways are ineffective as compared to upcoming research. Much research has been done in the domain of ML and DL to confront these challenges [1-6].
Within the framework of IoT, AI algorithms such as ML and DL techniques are working efficiently to cope with these obstacles [7-11]. Deep Learning, a subset of AI that can recognize nonlinear patterns more easily and helps in processing computing machines, is playing a key role in all this research and overcoming these challenges [12-16].
The fundamental contributions of our work are as follows:
● Advocating an AI Model-based approach using DL and ML techniques, along with ensemble learning classifiers for cyber attack detection in IoT, empowered by SMOTE(Synthetic Minority over-sampling technique)implementation [17].
● Enhancing the confidence and accuracy in detecting cyber security attacks in contrast to prevailing methods and approaches.
● Producing more precise results in detection by halting access to unauthorized systems, breaches, and interruptions.
● Resolving the class imbalance issue by SMOTE [18] to upgrade model generalization.
● Granting insights into the DL model with ensemble learning to detect and prevent cybersecurity attacks in IoT.
The research will contribute to and bring enhancement to the domain of cybersecurity and IoT security by employing ensemble learning. Moreover, it has the power to defend critical data and forecast upcoming attacks significantly.
Materials and Methods:
This study unveils an automated network detection model for IoT. It accumulates sensor flow data and utilizes feature engineering techniques such as balancing. In addition, DL models are tested for evaluating performance [24].
Bot-IoT Dataset:
A novel dataset named the Bot-IoT dataset has been used to accomplish this task. This dataset constitutes IoT data collected from the Cyber Range Lab of UNSW Canberra, incorporating normal traffic flows and flows generated by botnets emerging from several types of attacks.
A pragmatic testbed produced a detailed dataset with feature engineering to enhance the performance of machine learning models. Features were obtained from IoT services, network structure, and investigative analysis. Devices include smart fridges and smart doors, each performing specific functions.
The following are the target categories in the Bot-IoT dataset:
● Benign category: normal, legitimate IoT network activity without harmful Intent.
● UDP-focused DDoS attacks: flooding networks with TCP requests, exhibiting them impractical for authorized users.
● DDoS HTTP attacks: flooding the network with packets, making the service out of reach. ● DDoS HTTP attacks: flooding net server with HTTP requests leading to service disruption. ● TCP DoS attacks: leveraging TCP stack vulnerabilities to deplet devices.
● UDP DoS attacks: flooding targets with numerous packets causing service outages.
● HTTP-based DoS attacks: Overburdening the web servers with excessive requests causes underperformance.
● Keylogging: privately monitoring and recording keystrokes on compromised devices for harmful aims.
● Capture of data: unsanctioned capture of information from compromised IoT devices.
Table 1: Bot-IoT dataset.
| Type | Target | Count |
| BENIGN | Benign | 9543 |
| DDoS TCP | Attack | 19,547,603 |
| DDoS UDP | Attack | 18,965,106 |
| DDoS HTTP | Attack | 19,771 |
| DoS TCP | Attack | 12,315,997 |
| DoS UDP | Attack | 20,659,491 |
| DoS HTTP | Attack | 20,659,491 |
| Keylogging | Keylogging | 1469 |
| Data theft | Data theft | 118 |
| Total | – | 73,370,443 |
The Proposed Model
Data pre-processing:
The preprocessing of data is a significant part of developing the model. In the course of preprocessing, we dealt with missing values and cleaned the data through filtration, scaling, normalizing, and removing duplicate values. We used Python services while dealing with preprocessing [25-27].
Feature Engineering Techniques:
1. Correlation Coefficient:
The correlation coefficient quantifies the relationship between two factors in the dataset. In this context, the correlation coefficient highly helped in recognizing and analyzing the trends and patterns in the BoT IoT dataset. This dataset constitutes information about IoT devices that have been infiltrated by botnets. The correlation coefficient in this dataset has been determined for variables such as device type, network traffic pattern, communication protocols, and other connected factors. Once the variable has been selected, the correlation coefficient is measured by using certain statistical techniques that give insight into the data. A zero or near zero value indicates no relation, a high value indicates a direct relation, while a negative indicates an inversely proportional relation.
2. Feature Importance using Random Forest:
Feature significance analysis using Random Forest is an efficient technique that is significant for the BoT IoT dataset. This analysis discloses which factor has the greatest impression on the dependent variable. In this regard, the Bot-Iot dataset was first divided into training and testing datasets. The training dataset is for training the Random Forest model, while the testing subset is used for making predictions and evaluating model performance. Feature importance is measured using several techniques such as Gini impurity, entropy, and permutation importance. These techniques ascertain how each attribute is involved in reducing impurity in decision trees of random forests. The results can be represented using a bar chart and a pie chart. The results have been visualized in Figure 2. Applying feature importance analysis by using random forest certainly allows the authentication of key attributes directing the behaviour of IoT devices.
Many insignificant features such as pkSeqID’, ‘proto’, ‘saddr’, ‘sport’, ‘daddr’, ‘dport’, and ‘category have been dropped after applying the feature importance technique using Random Forest.
3. SMOTE Approach:
An elevated strategy for dealing with imbalanced data is the SMOTE approach, as shown in Figure 3. The SMOTE algorithm produces new samples by employing linear interpolation between the selection of the number of samples and those located in their neighbors [28,29].
To improve the classification impression of unbalanced data, a selected number of false samples has been produced as shown in Figure 4. This consequently transmutes an unbalanced data ratio into a balanced one.
Ensemble Learning:
1. Extra Trees Classifiers:
The Extra Tree classifier is an alternative to the random forest classifier. This Algorithm comprises more randomness in decision trees than in Random Forests [30]. It deploys ensemble learning in base trees to make decisions [31]. This technique is more useful when dealing with majority voting. In our specific context, the Extra Trees Classifier is being used to detect several types of attacks. This also aids in preventing the model from overfitting and improves the accuracy [32].
Figure 4: Attack class distributions after oversampling the BoT-IoT dataset using the SMOTE algorithm.
2. Histogram-based Gradient Boosting classifier:
This uses gradient-boosting, continuously improved training of an ensemble of weak learners [33]. This classifier uses a histogram to improve accuracy and predictive efficiency [34].
A histogram provides statistical information in the context of information distribution in each bin [35]. These measures enable effective calculation during training and also lower the loss rate, causing loss reduction. The learning rate parameter also assigns values to weak learners, therefore accomplishing the balance between model complexities and the ratio of convergence [36].
3. Adaptive Boosting classifier:
An Adaptive Boosting classifier(AdaBost) combines weak learners iteratively to form a stout classifier [37]. Decision trees act as weak learners, and their predictions weigh according to predictions. It provides more benefits to our computation such as being less prone to overfitting, and simple to implement, leading to a wealth of resources and expertise available to users. These all lead to more admiration of this algorithm in machine learning.
4. LGBM classifier:
The Light Gradient Boosting Machine(LGBM) is the most effective algorithm under gradient boosting framework [38]. It uses iterations to train weak learners while operating inside a gradient gradient boosting framework. Its renowned feature is its unique optimization that enhances its efficiency and improves accuracy.
5. CatBoosting classifier:
Catboosting is a very powerful technique of machine learning that is significant for classification. It falls under the category of gradient boosting structure that elevates the efficiency of the model [39]. It outperforms categorical data such as variables with high dimensions and cardinalities. It preserves the constancy of the model by randomly shuffling the order of each category during training, thereby uplifting the accuracy [40].
Evaluation Metrics:
The competency of the DL model is determined by using precision, recall, F1 score, computation time, and accuracy [41,42]. These matrices are computed with the help of TPR, TNR, FPR, and FPR.
● True positive rate (TPR): ratio of observed positives to expected positives
● False positive rate (FPR): ratio of values that are truly negative but are expected to be positive ● False negative rate (FNR): ratio values that are positive but observed to be negative ● True negative rate (TNR): ratio values that are negative but foreseen to become negative
● Precision: system’s capacity to accurately detect the presence of an attack; it shows the relationship between accurately predicted attacks and actual results:
Precision = TPR/(TPR + FPR)
● Recall: the system’s ability to correctly notice a botnet attack whenever occurs on a network: Recall = TPR/(TPR + FPR)
● Accuracy: the system’s ability to efficiently classify attack and non-attack packets Accuracy = (TNR + TPR)/(TPR + FNR + FPR + TNR)
● F1-score: average of recall and precision:
F1-Score = 2 × (Recall × Precision)/(Recall + Precision)
● Time complexity: how quickly or slowly an algorithm performs in the same relation to the amount of data.
Results:
Experimental Settings:
For this research, we used Python, TensorFlow, and Keras on GoogleColab GPU. Preprocessed attributes in the BoT_IoT dataset were used in the training procedure. We divided data into 70% training, 20% validation, and 10% testing. Elevation matrices, which have been discussed above, have been used for the evaluation of the model.
Experimental Results:
We performed 10 ML models for malware detection, including a single classifier, ensemble learning, and DL techniques. We contrasted the results with and without the SMOTE algorithm concisely used for managing imbalanced data.
Table 2: Performance results for detecting IoT network attacks without using the SMOTE algorithm.
| Metric | Accuracy | Precision | Recall | F1 score | CPU time | Model size(MB) |
| Random Forest | 0.9518 | 0.9538 | 0.9284 | 0.9403 | 21.6 s | 23.6 |
| Extra Trees | 0.9674 | 0.9652 | 0.9517 | 0.9582 | 47.6 s | 598.7 |
| KNN | 0.9083 | 0.9036 | 0.8869 | 0.8947 | 3.29 s | 13.6 |
| SVM | 0.6121 | 0.6280 | 0.3598 | 0.3695 | 21 min 50 s | 12.2 |
| HistGBoost | 0.9560 | 0.7488 | 0.7332 | 0.7321 | 13.4 s | 1.2 |
| AdaBoost | 0.1211 | 0.4552 | 0.3482 | 0.0826 | 1 min 19 s | 0.31 |
| LGBM | 0.9323 | 0.4665 | 0.4739 | 0.4690 | 36.1 s | 1.8 |
| CatBoost | 0.9819 | 0.9686 | 0.9608 | 0.9646 | 2 min 55 s | 3.5 |
| XGBoost | 0.9852 | 0.9806 | 0.9654 | 0.9727 | 2 min 43 s | 1.1 |
| MLP | 0.7539 | 0.3031 | 0.2942 | 0.2850 | 31.5 s | 0.005 |
| ANN | 0.8308 | 0.3308 | 0.4789 | 0.3701 | 13 min 48 s | 0.027 |
| LSTM | 0.7701 | 0.4887 | 0.3476 | 0.3682 | 10 min 10 s | 7.7 |
| GRU | 0.8536 | 0.6058 | 0.4517 | 0.4902 | 11 min 1 s | 7.7 |
| RNN | 0.8682 | 0.9189 | 0.7631 | 0.8013 | 10 min 50 s | 1.6 |
| Bagging | 0.9398 | 0.9324 | 0.9160 | 0.9238 | 2 min 54 s | 240.5 |
Table 3: Performance results for detecting IoT network attacks using the SMOTE algorithm from the BoT-IoT dataset.
| Metric | Accuracy | Precision | Recall | F1 score | CPU time | Model size(MB) |
| CatBoost | 0.97661 | 0.91249 | 0.9815 | 0.94369 | 7 min 43 s | 3.48 |
| XGBoost | 0.97986 | 0.94868 | 0.98084 | 0.96383 | 7 min 53 s | 1.22 |
| ANN | 0.76594 | 0.61794 | 0.89682 | 0.63602 | 31 min 41 s | 0.03 |
| MLP | 0.53336 | 0.31119 | 0.63571 | 0.32423 | 4 min 47 s | 0.02 |
| LSTM | 0.83418 | 0.75511 | 0.92699 | 0.76773 | 30 min 6 s | 7.69 |
| GRU | 0.87806 | 0.78463 | 0.93476 | 0.83175 | 29 min 50 s | 7.69 |
| RNN | 0.87147 | 0.77572 | 0.94066 | 0.8257 | 27 min 3 s | 1.62 |
| Bagging | 0.94099 | 0.91357 | 0.93127 | 0.92205 | 9 min 31 s | 350.73 |
| Random Forest | 0.9425 | 0.90961 | 0.9635 | 0.9304 | 1 min 7 s | 29.60 |
| Extra Trees | 0.90922 | 0.88756 | 0.8952 | 0.8906 | 3.43 s | 35.19 |
| KNN | 0.90922 | 0.88756 | 0.8952 | 0.8906 | 3.43 s | 35.19 |
| SVM | 0.59398 | 0.4853 | 0.63258 | 0.48259 | 1 h 18 min 19 s | 25.34 |
| HistGboost | 0.97437 | 0.97758 | 0.97437 | 0.97511 | 47.6 s | 1.90 |
| AdaBoost | 0.43068 | 0.32098 | 0.34041 | 0.25093 | 3 min 55 s | 0.31 |
| LGBM | 0.98242 | 0.96029 | 0.98055 | 0.96986 | 4 min 5 s | 11.05 |
Experiments without Using the SMOTE Algorithm:
The DL models show different results without SMOTE algorithm. Random Forest, KNN, and Extratrees exhibited fierce accuracy while SVM AdaBoosT accomplished lower accuracy specifically in imbalance data. Also, the model size varies, with ExtraTrees having the largest model. Overall RAndomForest and ExtraTrees outperformed without the SMOTE algorithm.
Figure 5 represents the examination of all models based on accuracy, precision, and F1-score.
Figure 6 shows the performance of DL models without using the SMOTE algorithm. MLP indicates a lower performance with an accuracy of 0.75 while, ANN performed better with improved accuracy and precision but failed while balancing data.
Figure 6: The proposed deep learning models’ evaluation results without using the SMOTE algorithm.
Proceeding to LSTM and GURU, both show equivalent results. LSTM achieves 0.78 accuracy while GURU achieves 0.85. Both accomplished precision better than ANN and MLP.
On the other hand, the RNN model outperforms all DL models and significantly achieves the highest accuracy. Nevertheless, it consumes the highest CPU time and largest model size as compared to others.
Experiments Using the SMOTE Algorithm:
Table 3 represents the performance results using the SMOTE algorithm. It contrasts the ML model performance based on precision and recall. Accuracy and F1 score for calculating impact for detecting attacks are shown in Figure 7.
CatBoost achieved 97.661% accuracy with a high precision of 0.91243 and a recall of 0.9815.
XGBoost accomplished 97.986% and precision much better than CatBoost 0.94886 and comparable recall of 0.98084. MLP model underperformed all others while ANN performed better than MLP but still trailed behind CatBoost and XGBoost.
The performance of LSTM, GRU, and RNN was seen as better than that of MLP and ANN architects, as shown in Figure 8. LSTM achieved an accuracy of 0.83418. GRU and RNN showed comparable accuracies of 0.878 and 0.87 respectively. However, all the models took higher CPU time and large model sizes.
CatBoost and XGBoost models outstripped all other models with high prescio, recall, and F1 scores. They consumed less training time and small model sizes. These traits lead them towards suitability for cybersecurity attack detection in IoT devices and networks.
Discussion:
Table 2 and Table 3 examine the classifier’s performance on the Bot-IoT dataset with and without the SMOTE algorithm. Figure 9 shows the ensemble learning with and without the SMOTE algorithm. Distinctively XGBoost and CatBoost are performing better than all other models and techniques.
XGBoost and CatBoost sustained higher accuracy, precision, recall, and F1 score. MLP, ANN, GRU, and LSTM didn’t perform well and showed lesser accuracy in handling imbalance data even with the SMOTE algorithm.
Classifiers with lower accuracies struggled hard to illustrate better performance. Among all, SVM with the least accuracy utilized comparatively greater CPU time showing its inefficiency in dealing with imbalanced data.
Recent research has utilized the DL models to enhance the efficiency level, but they don’t completely tackle the security needs.
Tale 4 shows and compares several cyber security solutions based on study year, dataset, methodology, algorithms, and accuracy results.
Table 4: Several cyber security solutions based on study year, dataset, methodology, algorithms, and accuracy results.
| Ref. | Dataset | Methodology | Accuracy(%) |
| Mendonça et al. [19] | DS2OS, CICIDS2017 | Deep learning | 98 |
| Popoola et al. [20] | BoT-IoT LAE | for dimensionality reduction and BLSTM classifier | 91.89 |
| Alharbi et al. [21] | N-BaIoT | A Local–Global best Bat Algorithm for Neural Networks | 90 |
| Saharkhizan et al. [22] | Modbus/TCP network traffic | LSTM and Ensemble learning | 98.99 |
| Pokhrel et al. [23] | BoT-IoT | Deep learning | 87.4 |
| Proposed | BoT-IoT | CatBoosting XGBoosting | 98.19 98.52 |
Conclusions:
This research reveals a better and smart way for detecting malware attacks in IoT networks. It integrates DL models with other techniques to avoid overfitting. CatBoost and XGBoost are best at spotting attacks on IoT networks. Future work will include different methods and techniques for detecting malware attacks.
References:
1. Vermesan, O.; Friess, P.; Guillemin, P.; Giaffreda, R.; Grindvoll, H.; Eisenhauer, M.; Serrano, M.; Moessner, K.; Spirito, M.; Blystad, L.-C. Internet of Things beyond the Hype: Research, Innovation and Deployment. In Building the Hyperconnected Society-Internet of Things Research and Innovation Value Chains, Ecosystems and Markets; River Publishers: Gistrup, Denmark, 2022; pp. 15–118; ISBN 1-00-333745-7.
2. Madina, S.F.; Islam, M.S.; Alamgir, F.M.; Ferdous, M.F. Internet of Things (IoT)-Based Industrial Monitoring System. In Industrial Internet of Things; CRC Press: Boca Raton, FL, USA, 2022; pp. 55–86; ISBN 1-00-310226-3.
3. Huang, L. Design of an IoT DDoS Attack Prediction System Based on Data Mining Technology. J. Supercomput. 2022, 78, 4601–4623. [CrossRef]
4. Krichen, M. A Survey on Formal Verification and Validation Techniques for Internet of Things. Appl. Sci. 2023, 13, 8122. [CrossRef]
5. Idrissi, I.; Azizi, M.; Moussaoui, O. IoT Security with Deep Learning-Based Intrusion Detection Systems: A Systematic Literature Review. In Proceedings of the 2020 Fourth International Conference on Intelligent Computing in Data Sciences (ICDS), Fez, Morocco, 21–23 October 2020; IEEE: Piscataway, NJ, USA, 2020; pp. 1–10.
6. Krichen, M. Strengthening the Security of Smart Contracts through the Power of Artificial Intelligence. Computers 2023, 12, 107. [CrossRef]
7. Abdalzaher, M.S.; Krichen, M.; Yiltas-Kaplan, D.; Ben Dhaou, I.; Adoni, W.Y.H. Early Detection of Earthquakes Using IoT and Cloud Infrastructure: A Survey. Sustainability 2023, 15, 11713. [CrossRef]
8. Majid, M.; Habib, S.; Javed, A.R.; Rizwan, M.; Srivastava, G.; Gadekallu, T.R.; Lin, J.C.-W. Applications of Wireless Sensor Networks and Internet of Things Frameworks in the Industry Revolution 4.0: A Systematic Literature Review. Sensors 2022, 22, 2087. [CrossRef]
9. Oseni, A.; Moustafa, N.; Creech, G.; Sohrabi, N.; Strelzoff, A.; Tari, Z.; Linkov, I. An Explainable Deep Learning Framework for Resilient Intrusion Detection in IoT-Enabled Transportation Networks. IEEE Trans. Intell. Transp. Syst. 2022, 24, 1000–1014. [CrossRef]
10. Nasir, M.; Javed, A.R.; Tariq, M.A.; Asim, M.; Baker, T. Feature Engineering and Deep Learning Based Intrusion Detection Framework for Securing Edge IoT. J. Supercomput. 2022, 78, 8852– 8866. [CrossRef]
11. Baduge, S.K.; Thilakarathna, S.; Perera, J.S.; Arashpour, M.; Sharafi, P.; Teodosio, B.; Shringi, A.; Mendis, P. Artificial Intelligence and Smart Vision for Building and Construction 4.0: Machine and Deep Learning Methods and Applications. Autom. Constr. 2022, 141, 104440. [CrossRef]
12. Boulila, W.; Driss, M.; Alshanqiti, E.; Al-Sarem, M.; Saeed, F.; Krichen, M. Weight Initialization Techniques for Deep Learning Algorithms in Remote Sensing: Recent Trends and Future Perspectives. In Advances on Smart and Soft Computing; Saeed, F., Al-Hadhrami, T., Mohammed, E., Al-Sarem, M., Eds.; Springer: Singapore, 2022; pp. 477–484.
13. Islam, U.; Muhammad, A.; Mansoor, R.; Hossain, M.S.; Ahmad, I.; Eldin, E.T.; Khan, J.A.; Rehman, A.U.; Shafiq, M. Detection of Distributed Denial of Service (DDoS) Attacks in IOT Based Monitoring System of Banking Sector Using Machine Learning Models. Sustainability 2022, 14, 8374. [CrossRef]
14. Vadivelan, N.; Bhargavi, K.; Kodati, S.; Nalini, M. Detection of Cyber Attacks Using Machine Learning. In AIP Conference Proceedings; AIP Publishing LLC: Melville, NY, USA, 2022; Volume 2405, p. 030003.
15. Ahmad, R.; Alsmadi, I.; Alhamdani, W.; Tawalbeh, L. A Comprehensive Deep Learning Benchmark for IoT IDS. Comput. Secure. 2022, 114, 102588. [CrossRef]
16. Iwendi, C.; Rehman, S.U.; Javed, A.R.; Khan, S.; Srivastava, G. Sustainable Security for the Internet of Things Using Artificial Intelligence Architectures. ACM Trans. Internet Technol. 2021, 21, 1–22. [CrossRef]
17. Fernández, A.; Garcia, S.; Herrera, F.; Chawla, N.V. SMOTE for Learning from Imbalanced Data: Progress and Challenges, Marking the 15-Year Anniversary. J. Artif. Intell. Res. 2018, 61, 63–905. Available online: https://www.jair.org/index.php/jair/ article/view/11192 (accessed on 24 September 2023). [CrossRef]
18. Torgo, L.; Ribeiro, R.P.; Pfahringer, B.; Branco, P. SMOTE for Regression. In Progress in Artificial Intelligence; Correia, L., Reis, L.P., Cascalho, J., Eds.; Springer: Berlin/Heidelberg, Germany, 2013; pp. 378–389.
19. Mendonça, R.V.; Silva, J.C.; Rosa, R.L.; Saadi, M.; Rodriguez, D.Z.; Farouk, A. A Lightweight Intelligent Intrusion Detection System for Industrial Internet of Things Using Deep Learning Algorithms. Expert Syst. 2022, 39, e12917. [CrossRef]
20. Popoola, S.I.; Adebisi, B.; Hammoudeh, M.; Gui, G.; Gacanin, H. Hybrid Deep Learning for Botnet Attack Detection in the Internet-of-Things Networks. IEEE Internet Things J. 2020, 8, 4944–4956. [CrossRef]
21. Alharbi, A.; Alosaimi, W.; Alyami, H.; Rauf, H.T.; Damaševiˇcius, R. Botnet Attack Detection Using Local Global Best Bat Algorithm for Industrial Internet of Things. Electronics 2021, 10, 1341. [CrossRef]
22. Saharkhizan, M.; Azmoodeh, A.; Dehghantanha, A.; Choo, K.-K.R.; Parizi, R.M. An Ensemble of Deep Recurrent Neural Networks for Detecting IoT Cyber Attacks Using Network Traffic. IEEE Internet Things J. 2020, 7, 8852–8859. [CrossRef]
23. Pokhrel, S.; Abbas, R.; Aryal, B. IoT Security: Botnet Detection in IoT Using Machine Learning. arXiv 2021, arXiv:2104.02231.
24. Garavand, A.; Behmanesh, A.; Aslani, N.; Sadeghsalehi, H.; Ghaderzadeh, M. Towards Diagnostic Aided Systems in Coronary Artery Disease Detection: A Comprehensive Multiview Survey of the State of the Art. Int. J. Intell. Syst. 2023, 2023, 6442756. Available online:
https://www.hindawi.com/journals/ijis/2023/6442756/ (accessed on 24 September 2023). [CrossRef]
25. Fan, C.; Chen, M.; Wang, X.; Wang, J.; Huang, B. A Review on Data Preprocessing Techniques toward Efficient and Reliable Knowledge Discovery From Building Operational Data. Front. Energy Res. 2021, 9, 652801. [CrossRef]
26. Ghaderzadeh, M.; Aria, M.; Asadi, F. X-Ray Equipped with Artificial Intelligence: Changing the COVID-19 Diagnostic Paradigm during the Pandemic. BioMed Res. Int. 2021, 2021, e9942873. [CrossRef]
27. Chawla, N.V.; Bowyer, K.W.; Hall, L.O.; Kegelmeyer, W.P. SMOTE: Synthetic Minority over-Sampling Technique. J. Artif. Intell. Res. 2002, 16, 321–357. [CrossRef]
28. Nagisetty, A.; Gupta, G.P. Framework for Detection of Malicious Activities in IoT Networks Using Keras Deep Learning Library. In Proceedings of the 2019 3rd International Conference on
Computing Methodologies and Communication (ICCMC), Erode, India, 27–29 March 2019; IEEE: Piscataway, NJ, USA, 2019; pp. 633–637.
29. González, S.; García, S.; Del Ser, J.; Rokach, L.; Herrera, F. A Practical Tutorial on Bagging and Boosting Based Ensembles for Machine Learning: Algorithms, Software Tools, Performance Study, Practical Perspectives and Opportunities. Inf. Fusion 2020, 64, 205–237. [CrossRef]
30. Acosta, M.R.C.; Ahmed, S.; Garcia, C.E.; Koo, I. Extremely Randomized Trees-Based Scheme for Stealthy Cyber-Attack Detection in Smart Grid Networks. IEEE Access 2020, 8, 19921–19933. [CrossRef]
31. Seyghaly, R.; Garcia, J.; Masip-Bruin, X.; Varnamkhasti, M.M. Interference Recognition for Fog Enabled IoT Architecture Using a Novel Tree-Based Method. In Proceedings of the 2022 IEEE International Conference on Omni-Layer Intelligent Systems (COINS), Barcelona, Spain, 1–3 August 2022; IEEE: Piscataway, NJ, USA, 2022; pp. 1–6.
32. Alghamdi, R.; Bellaiche, M. Evaluation and Selection Models for Ensemble Intrusion Detection Systems in IoT. IoT 2022, 3, 285–314. [CrossRef]
33. Almomani, O.; Almaiah, M.A.; Alsaaidah, A.; Smadi, S.; Mohammad, A.H.; Althunibat, A. Machine Learning Classifiers for Network Intrusion Detection System: Comparative Study. In Proceedings of the 2021 International Conference on Information Technology (ICIT), Amman, Jordan, 14–15 July 2021; IEEE: Piscataway, NJ, USA, 2021; pp. 440–445.
34. Heinrich, C. On the Number of Bins in a Rank Histogram. Q. J. R. Meteorol. Soc. 2021, 147, 544– 556. [CrossRef]
35. Wang, K.; Lu, J.; Liu, A.; Zhang, G.; Xiong, L. Evolving Gradient Boost: A Pruning Scheme Based on Loss Improvement Ratio for Learning under Concept Drift. IEEE Trans. Cybern. 2021, 53, 2110– 2123. [CrossRef] [PubMed]
36. Ding, Y.; Zhu, H.; Chen, R.; Li, R. An Efficient AdaBoost Algorithm with the Multiple Thresholds Classification. Appl. Sci. 2022, 12, 5872. [CrossRef]
37. Mishra, D.; Naik, B.; Nayak, J.; Souri, A.; Dash, P.B.; Vimal, S. Light Gradient Boosting Machine with Optimized Hyperparameters for Identification of Malicious Access in IoT Network. Digit. Commun. Netw. 2023, 9, 125–137. [CrossRef]
38. Sanjeetha, R.; Raj, A.; Saivenu, K.; Ahmed, M.I.; Sathvik, B.; Kanavalli, A. Detection and Mitigation of Botnet Based DDoS Attacks Using Catboost Machine Learning Algorithm in SDN Environment. Int. J. Adv. Technol. Eng. Explor. 2021, 8, 445.
39. Prokhorenkova, L.; Gusev, G.; Vorobev, A.; Dorogush, A.V.; Gulin, A. CatBoost: Unbiased Boosting with Categorical Features. In Advances in Neural Information Processing Systems; Neural Information Processing Systems Foundation, Inc. (NeurIPS): La Jolla, CA, USA, 2018; Volume 31.
40. Ghaderzadeh, M.; Aria, M.; Hosseini, A.; Asadi, F.; Bashash, D.; Abolghasemi, H. A Fast and Efficient CNN Model for B-ALL Diagnosis and Its Subtypes Classification Using Peripheral Blood Smear Images. Int. J. Intell. Syst. 2022, 37, 5113–5133. Available online:
https://onlinelibrary.wiley.com/doi/abs/10.1002/int.22753 (accessed on 24 September 2023). [CrossRef]
41. Hosseini, A.; Eshraghi, M.A.; Taami, T.; Sadeghsalehi, H.; Hoseinzadeh, Z.; Ghaderzadeh, M.; Rafiee, M. A Mobile Application Based on Efficient Lightweight CNN Model for Classification of B ALL Cancer from Non-Cancerous Cells: A Design and Implementation Study. Inform. Med. Unlocked 2023, 39, 101244. [CrossRef]
Article By: Sulaikha Fiyas
