DETECTING AND EVALUATING FAKE WEBSITES USING PATTERN RECOGNITION ALGORITHMS

174 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ DETECTING AND EVALUATING FAKE WEBSITES USING PATTERN RECOGNITION ALGORITHMS Kabulov Anvar, Professor of the National University of Uzbekistan named after Mirzo Ulugbek [email protected] Otakhanov Alisherbek, Doctoral student of Fergana State University [email protected] Abstract: The increase in the creation of fake web pages by attackers is leading to a sharp increase in cyberattacks. Attackers use these fake web sites to advertise products to Internet users, distribute malicious programs, or steal users' valuable logins and passwords. Traditional solutions for detecting such fake web addresses are not effective in detecting newly created fake web addresses. In this article, we propose a new approach that combines several machine learning algorithms. Therefore, we use various selected features to improve the accuracy of sorting and classifying web pages. From our experimental results, it can be seen that using the proposed approach, the Random Forest (RF) classifier showed the best accuracy of 99%. It can be seen that the Random Forest classifier can be considered more reliable than the others in detecting fake web addresses. Keywords: web Security, Machine Learning, Random Forest, Cyberattacks, Fake Websites, URL. Introduction. The rapid development, expansion, and increased accessibility of the Internet have led consumers to shift from traditional shopping to online shopping. However, this innovation has also increased cyber threats. Attackers use strategies such as fake websites to steal confidential account information from unsuspecting users. Fraudsters use fake websites to trick individuals into revealing sensitive information, including credit card numbers, passwords, and personal information [1]. The creation of fake websites has attracted significant attention from security researchers due to its potential to exploit users. Although users are able to detect these fake pages by carefully examining URLs, the busyness of online activities sometimes causes them to overlook such differences, which can lead them to fall prey to attackers. Fake websites affect trust in online shopping and cause financial losses to individuals. As a Verizon data breach study shows, the initial step of accessing a fake website is responsible for 90 percent of all frauds [2]. In addition to obtaining personal and confidential information, the purpose of modern fake websites is to infect victims’ computers with various forms of malware [3], [4]. Communication channels such as the Internet, SMS, and email are used to distribute these fake websites. The Internet serves as a means for attackers to communicate with victims through email messages, fake websites, instant messaging, and social networks [5], [6]. In this study, we propose a comprehensive approach to detect and prevent fraudulent websites, which includes the use of machine learning methods. Thus, this research work allows us to more accurately identify such websites. Our main goal is to develop a system with enhanced capabilities for detecting and classifying fraudulent websites and then preventing them. Using machine learning algorithms, we extract 27 features from websites, including URL and domainbased attributes, and attributes based on the directory and parameter parts of the URL. The proposed approach is aimed at strengthening the security of Internet users and protecting them from the breach of their personal and financial data. To evaluate the effectiveness of our approach, we perform a comparative analysis of 7 machine learning models. 175 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ These are: Naive Bayes, Decision Tree, K-nearest Neighbor, Support Vector Machine, Random Forest, Logistic Regression, and Adaboost. This research will provide an opportunity to improve website security and protect confidential information from malicious attacks. In this research, we aim to increase knowledge about how to avoid being fooled by fake websites and distinguish them from real ones. Related works. This section reviews various methods for detecting fake websites using machine learning algorithms and website features. Chiew and Chang [7] proposed a method that relies on website logos. They extracted the logo from a web page and fed it to Google’s image search engine to learn how to detect suspicious websites. They were able to determine whether a website was legitimate or fake by comparing the website with search engine results. They present a content-agnostic method for predicting website domains based on certificate transparency logs and passive DNS records. The study demonstrates the usefulness of this analysis by training a classifier with unique features and achieving low false positive rates, as well as high accuracy and recall in predicting fake website domains [8]. Heuristic solutions for detecting fake websites study various features such as non-content-based, content-based, and visual similarity-based features, as well as DNS information and the registries used to register the site [9]. The paper proposes a general solution based on heuristics for these features to predict fake websites. Sinha [10] proposes a dataset of 198 features from fake websites as an empirical approach to detecting fake websites. The dataset is analyzed using machine learning and deep learning models, and Random Forest detects fake web pages with high accuracy. The paper emphasizes the importance of obtaining a diverse set of features for effective detection of fake websites. The use of machine learning to detect fake websites and to highlight the limitations of blacklists in detecting real-time attacks [11]. The study focused on fake website URLs and domains associated with Italian organizations, and models based on pre-trained encoders and convolutional neural networks showed promising results. In this paper, we present a method for detecting and preventing fake websites. We extract a set of attributes from their URL address and then classify them as legitimate or fake websites. Specifically, we rely on the use of machine learning models to detect potential fake websites based on their URL address. We then derive a complete set of 27 features, including URL-based features, domain-based features, URLbased features, and URL-based features and directorybased features, to further evaluate a website and determine its authenticity. We compared different machine learning models to evaluate the effectiveness of our proposed approach. In this paper, websites were evaluated based on URL features using Naive Bayes (NB), Decision Tree (DT), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and Adaboost algorithms. Our approach shows that the proposed approach is effective in detecting fake websites with a very good percentage. The proposed method. Our proposed approach extracts and analyzes various features from URLs to effectively detect fake websites. The main contribution of this paper is that the extracted feature set is used together. We build a knowledge base from the extracted features. Based on this knowledge base, we check and predict whether the URLs of other web pages are real or fake. We propose the use of a rule set to improve the accuracy of detecting fake web pages. Architecture of the proposed method. The general architecture of the proposed approach is divided into three stages. In the first stage, all the important features of the URLs are extracted. In the second stage, a knowledge base is formed based on the production logic. In the third stage, the extracted web pages are tested to determine whether they are fake or genuine. Fig. 1 illustrates the architecture of the proposed approach. The details of each stage in the architecture are described below. 176 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ Figure 1. General architecture of the proposed approach. Create features. The feature set is created as follows. Our features are based on the URL of a web page. We extract the features from the URL using software (C#). We divide the features into four groups according to the location where they are detected, as shown in Table 1. Specifically, features 𝑓1,…,𝑓9 belong to the general URL, features 𝑓10,…,𝑓13 belong to the domain part of the URL, features 𝑓14,…,𝑓17 belong to the directory part of the URL, and features 𝑓18,…,𝑓27 belong to the parameter part of the URL. A detailed explanation of the detected features is given in the feature extraction section of this article. Table 1. Features used in the proposed approach Categor y Sign Name of Feature URLbased features 𝑓1,𝑓2,𝑓3,𝑓4,𝑓5, 𝑓6,𝑓7,𝑓8,𝑓9 count_dot_url, count_slash_url, count_tld_url, length_url, count_at_url, count_hyphen_url, count_underline_url, count_equal_url, count_and_url Domain -based features 𝑓10,𝑓11,𝑓12,𝑓13 count_dot_domain, count_vowels_domain , domain_length, count_hyphen_domain Director y-based features 𝑓14,𝑓15,𝑓16,𝑓17 count_slash_directory, directory_length, count_dot_directory, count_hyphen_directo ry Paramet er-based features 𝑓18,𝑓19,𝑓20,𝑓21,𝑓22, 𝑓23,𝑓24,𝑓25,𝑓26,𝑓27 count_hyphen_params , count_at_params, count_dot_params, count_equal_params, count_and_params, params_length, count_params, count_underline_para ms, count_slash_params, count_questionmark_p arams Vectorization of features. After feature extraction, feature vectorization is applied to create a feature vector for each URL. After that, a structured database is created. We divide the URL features into 4 groups and create the feature vector required to train the proposed approach. The combination of 9 features belonging to the general URL address forms a 9dimensional feature vector 𝐹𝑈=〈𝑓1,𝑓2,𝑓3,…,𝑓9〉, while the combination of 4 features belonging to the domain part of the URL forms a 4-dimensional feature vector 𝐹𝐷=〈𝑓10,𝑓11,𝑓12,𝑓13〉. The combination of 4 features belonging to the directory part of the web page URL produces a 4-dimensional feature vector 𝐹𝐶= 〈𝑓14,𝑓15,𝑓16,𝑓17〉, and finally, the combination of 10 features belonging to the parameter part of the URL produces a 10-dimensional feature vector 𝐹𝑃= 〈𝑓18,𝑓19,𝑓20,…,𝑓27〉. The above 4 feature vectors are combined to produce the final feature vector 𝐹𝑉= 𝐹𝑈⋃𝐹𝐷⋃𝐹𝐶⋃𝐹𝑃=〈𝑓1,𝑓2,𝑓3,𝑓4,𝑓5,…,𝑓27〉, and this feature vector is given as input to machine learning algorithms, which then classify the website. Detection module. The detection module involves building a powerful classifier using machine learning classifiers. As a result, the classifier is effective in detecting fake websites. The classifiers we use here apply features based on the total length of the URL 𝐹𝑈, features based on the domain part of the URL 177 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ 𝐹𝐷, features related to the directory part 𝐹𝐶, and features based on the parameter part 𝐹𝑃 to a combined set of features. In the training phase, the classifiers are trained using the feature vector 𝐹𝑈⋃𝐹𝐷⋃𝐹𝐶⋃𝐹𝑃 collected from each record in the training dataset. In the testing phase, the classifiers determine whether a given website is a fake or a real website. A detailed description is shown in Fig. 2. Figure 2. Algorithm to detect fake website. The novelty of this methodology is the use of machine learning techniques to achieve high accuracy and efficiency in detecting fake websites. Machine learning techniques provide a powerful and flexible way to deal with new or unknown fake websites. This methodology can improve web security and protect users from fake web addresses. Extracting features. We used 27 features to categorize websites, including URL-based features, domain-based features, directory-based features, and parameter-based features. Table 1 shows the features used in this study. Mathematical and algorithmic formalization and evaluation methods. This section presents the algorithms used in the article and their mathematical formalization. In addition, information is provided about the machine learning algorithms used in the proposed model and their evaluation methods. Mathematical formalization based on Petri nets. The mathematical formalization [12] of a threestage machine learning system based on Petri nets is construct on the fundamental description of Petri nets: 𝑃𝑁=(𝑃,𝑇,𝐹,𝑊,𝑀0) (1) here, 𝑃={𝑝1,𝑝2,…,𝑝9} – positions, 𝑇= {𝑡1,𝑡2,…,𝑡7} – transitions, 𝐹⊆(𝑃×𝑇)∪(𝑇×𝑃) – flow relation, 𝑊:𝐹→𝑁+ – weights, 𝜇0:𝑃→𝑁 – initial marking. The positions and transitions in a Petri net [13] are named as follows: Positions: − 𝑝1: incoming URLs − 𝑝2: the process of feature extraction − 𝑝3: vectorized features − 𝑝4: knowledge base − 𝑝5: model training stage − 𝑝6: model testing stage − 𝑝7: model analysis − 𝑝8: result (Legitimate or Fake) − 𝑝9: safety decision making Transitions: − 𝑡1: extract features − 𝑡2: vectorize − 𝑡3: generate knowledge base − 𝑡4: train ML models − 𝑡5: input test URLs − 𝑡6: classify using ML algorithms − 𝑡7: security decision Using the above positions and transitions, the 𝐹-flow relationship is formed as follows: 𝐹= { (𝑝1,𝑡1),(𝑡1,𝑝2), (𝑝2,𝑡2),(𝑡2,𝑝3), (𝑝3,𝑡3),(𝑡3,𝑝4), (𝑝4,𝑡4),(𝑡4,𝑝5), (𝑝6,𝑡5),(𝑡5,𝑝6), (𝑝5,𝑡6),(𝑝6,𝑡6),(𝑡6,𝑝7), (𝑝7,𝑡7),(𝑡7,𝑝8),(𝑡7,𝑝9) } Based on the F – flow relationship, the active transitions [14] in Table 2 were generated. Table 2. Workflow table of active transitions from previous state to new state Station Active transitio n New station 𝜇0 =(1,0,0,0,0,1,0,0,0) 𝑡1 𝜇1 =(0,1,0,0,0,1,0,0,0) 𝜇1 =(0,1,0,0,0,1,0,0,0) 𝑡2 𝜇2 =(0,0,1,0,0,1,0,0,0) 𝜇2 =(0,0,1,0,0,1,0,0,0) 𝑡3 𝜇3 =(0,0,0,1,0,1,0,0,0) 𝜇3 =(0,0,0,1,0,1,0,0,0) 𝑡4 𝜇4 =(0,0,0,0,1,1,0,0,0) 178 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ 𝜇4 =(0,0,0,0,1,1,0,0,0) 𝑡5 𝜇5 =(0,0,0,0,1,0,1,0,0) 𝜇5 =(0,0,0,0,1,0,1,0,0) 𝑡6 𝜇6 =(0,0,0,0,0,0,0,1,0) 𝜇6 =(0,0,0,0,0,0,0,1,0) 𝑡7 𝜇7 =(0,0,0,0,0,0,0,0,1) The order of operation of the Petri net is described by the following stages of transitions: 𝜎=𝑡1∘𝑡2∘𝑡3∘𝑡4∘𝑡5∘𝑡6∘𝑡7 Machine learning algorithms used in the article. In this study, we compared the performance of 7 classifiers used as machine learning methods for the proposed system: Naive Bayes (NB), Decision Tree (DT), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and Adaboost. About the dataset used in the article. The proposed method for classifying URLs as legitimate and fake web pages was tested using a dataset. The dataset used in the tests contains URLs of 6000 web pages, some of which are used for fake web pages and others are legitimate. We use the database of fake and real websites on the Kaggle open data platform. We extract 27 attributes for each website that is part of the dataset. The list includes various features such as URL length, number of special characters in the URL, and features related to the domain, parameter, and directory part of the URL. We summarize the features of the dataset used for the experiments and evaluation in Table 1. We assign a value of 1 to each feature in the dataset if it is a fake website and a value of 0 if it is a legitimate website. Evaluation parameters used. In our experiment, we compared different classifiers using evaluation metrics such as false positive rate, false negative rate, precision, recall, F1-score, and accuracy to evaluate the performance of the proposed system. According to Table 4, these parameters are calculated using the True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) fields of the confusion matrix shown in Table 3. Table 3.Confusion matrix Class Fake Legitimate Fake TP FP Legitimate FN TN FPR (False Positive Rate): This is the percentage of false detection of legitimate websites, which is calculated according to (3): 𝐹𝑃𝑅=𝐹𝑃 𝐹𝑃+𝑇𝑁 (2) FNR (False Negative Rate): This is the percentage of falsely classified fake websites, calculated as follows (4): 𝐹𝑁𝑅= 𝐹𝑁 𝑇𝑃+𝐹𝑁 (3) Precision: It assesses how accurate the model is. It is possible to correctly classify the true result, and it is calculated as follows (5): 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛= 𝑇𝑃 𝑇𝑃+𝐹𝑃 (4) Recall: The ability of a model to correctly predict positives from true positives is indicated by the model's recall score, which is calculated by (6): 𝑅𝑒𝑐𝑎𝑙𝑙= 𝑇𝑃 𝐹𝑁+𝑇𝑃 (5) F-measure: This is similar to the accuracy and harmonic mean. It provides a fast way to compare classifiers and is between 0 and 1, calculated by (7): 𝐹−𝑚𝑒𝑎𝑠𝑢𝑟𝑒= 2∗𝑇𝑃 2∗𝑇𝑃+𝐹𝑁+𝐹𝑃 (6) Accuracy (%): This is a well-defined percentage of legitimate and fake websites, calculated according to (8): 𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦= 𝑇𝑃+𝑇𝑁 𝑇𝑃+𝐹𝑁+𝑇𝑁+𝐹𝑃∗100 (7) Table 4. Fields of the confusion matrix and their definition Field Definition TP Number of websites correctly identified as fake TN Number of websites identified as legitimate FP Number of legitimate websites incorrectly identified as fake websites FN Number of fake websites incorrectly identified as legitimate These fields are used to calculate a number of performance metrics, including precision (%), accuracy, recall, and F1 score. We use these metrics to evaluate the performance of the algorithm and then make necessary changes to improve it. 179 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ Results and discussion. Classification results using features (such as domain-based features, parameter-based features, URL-based features, and directory-based features) are shown in this section. According to equations (2) to (7), the accuracy (%), precision, recall, F1 score, and FPR and FNR values of the recommended features are shown in Table 5. Seven classifiers were used to classify fake websites according to feature groups. Using the data in this table, we can understand how each feature affects the classification. The findings show that the RF algorithm provides the most accurate classification with the lowest FPR and FNR. Table 5. Performance evaluation results of different classifiers The confusion matrix for the 7 classifier algorithms we used in our study is shown in Fig. 3. As can be seen from this figure, the most efficient algorithm was the random forest classifier. The algorithm that achieved the lowest accuracy was the Naïve Bayes classifier. Figure 3. Confusion matrices of algorithms. The performance of the recommended feature set was evaluated and the results were summarized for the seven classifiers considered based on accuracy. Fig. 4 shows that RF with an accuracy of 99% and DT with an accuracy of 98.17% both achieved the maximum accuracy for this feature set. Figure 4. Accuracy (%) results of algorithms. Accuracy is defined as the ratio of true positive predictions to all positive predictions (including true positive and false positive predictions). Therefore, the amount of false positive predictions decreases as the accuracy increases. In this case, RF and DT algorithms have the highest accuracy scores, i.e., 99.22% and 98.33%, respectively. The results are shown in Fig. 5. Figure 5. Precision results of algorithms. 180 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ The results in Fig. 6 show the recall values for various machine learning algorithms used to classify websites into legitimate and fake websites. It measures how well the model can distinguish between positive cases (fake websites) and true positive examples. Figure 6. Recall results of algorithms. The F1 scores for various machine learning algorithms for classifying websites as legitimate or fake are shown in Fig. 7. The RF algorithm is the most accurate method for classifying websites as legitimate or fake. Figure 7. Evaluation of algorithm results using F1 score. Fig. 8 below shows a comparison of the results of Accuracy (%), Precision, Recall, and F1 score for seven algorithms used to detect fake websites. In this case, it shows that the random forest algorithm is the best. Figure 8. Comparison of evaluation algorithms. In Fig. 9, we selected the important features of the above 3 models (DT, RF, AdaBoost) by mutual voting in order of decreasing common important features for the proposed model. Figure 9. The most important features of the 3 models (DT, RF, AdaBoost). Conclusion. In this study, we presented a powerful method for detecting fake websites using machine learning techniques. We used URL features of a web page, domain-based features, and features related to the directory and parameter parts of the URL to inspect the web page. The results of our research show that the RF classifier achieved the highest accuracy of 99%, with FPR and FNR rates of 0.78 and 181 “Al-Farg‘oniy avlodlari” elektron ilmiy jurnali ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil "Descendants of Al-Farghani" electronic scientific journal. ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 year Электронный научный журнал "Потомки АльФаргани" ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год https://al-fargoniy.uz/ 1.22, respectively. In addition, the decision tree model demonstrated excellent performance with an accuracy of 98.17%, despite a slight increase in FPR and FNR rates of 1.67 and 2.00. Countering the sophisticated strategies of creating fake websites requires a proactive approach. As cybercriminals constantly improve their methods, it becomes necessary to develop increasingly powerful and effective systems. Our study is one of the important steps towards solving this problem. References 1. Singla, S., Gandotra, E., Bansal, D., & Sofat, S. “A novel approach to malware detection using static classification”, International Journal of Computer Science and Information, Vol.13, No.3, pp.1-5, 2015. 2. Enterprise, V. "Verizon 2018 data breach investigations report", 2018. [Online]. Available: https://verizon.com/business/resources/reports/2018 - data-breach-digest.pdf 3. Gandotra, E., Bansal, D., & Sofat, S. “Malware intelligence: beyond malware analysis”, International Journal of Advanced Intelligence Paradigms, Vol.13, No.1-2, pp.80-100, 2019. DOI: 10.1504/IJAIP.2019.099945 4. Sharma, A., Gandotra, E., Bansal, D., & Gupta, D. “Malware capability assessment using fuzzy logic”, Cybernetics and Systems, Vol.50, No.4, pp.323-338, 2019. DOI: 10.1080/01969722.2018.1552906 5. Chiew, K.L., Yong, K.S.C., & Tan, C.L.J.E.S.w.A. “A survey of phishing attacks: Their types, vectors, and technical approaches”, Expert Systems with Applications, Vol.106, pp.1-20, 2018. DOI: 10.1016/j.eswa.2018.03.050 6. Gandotra, E., & Sofat, S.J.I.J.o.N.-G.C. “Tools & Techniques for Malware Analysis and Classification”, International Journal of NextGeneration Computing, Vol.7, No.3, pp.176-197, 2016. 7. Chiew, K. L., Chang, E. H., & Tiong, W. K. “Utilization of website logo for phishing detection”, Computers & Security, Vol.54, pp.16-26, 2015. DOI: 10.1016/j.cose.2015.07.006 8. AlSabah, M., Nabeel, M., Boshmaf, Y., & Choo, E. “Content-Agnostic Detection of Phishing Domains Using Certificate Transparency and Passive DNS”, Proceedings of the 25th International Symposium on Research in Attacks, Intrusions, and Defenses, pp. 446-459, 2022. DOI: 10.1145/3545948.3545958 9. Torrealba A, L. and Bustos-Jiménez, J. “Detecting Phishing in a Heuristic Way (Abstract)”, 2021. 10. Sinha, J., & Sachan, M. “PhishX: An Empirical Approach to Phishing Detection”, 2022. DOI: 10.1145/1122445.1122456 11. Ranaldi, L., Petito, M., Gerardi, M., Fallucchid, F., & Zanzotto, F.M. “Machine Learning Techniques for Italian Phishing Detection”, in Italian Conference on Cybersecurity, Rome, Italy 2022. 12. A.Kabulov, I. Yarashov and A. Otakhonov, "Algorithmic Analysis of the System Based on the Functioning Table and Information Security," 2022 IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS), Toronto, ON, Canada, 2022, pp. 1-5, doi: 10.1109/IEMTRONICS55184.2022.9795746. 13. Normatov, I., Yarashov, I., Otakhonov, A., & Ergashev, B. (2022, September). Construction of reliable well distribution functions based on the principle of invariance for convenient user access control. In 2022 International Conference on Information Science and Communications Technologies (ICISCT) (pp. 1-5). IEEE. 14. Toshmatov, S., Yarashov, I., Otakhonov, A., & Ismatillayev, A. (2022, September). Designing an algorithmic formalization of threat actions based on a Functioning table. In 2022 International Conference on Information Science and Communications Technologies (ICISCT) (pp. 1-5). IEEE.