Instagram fake profile detection using an ensemble learning method

May combine two claims.

Social media platforms have implemented authentication systems and detection algorithms, but the problem of fake accounts remains persistent.

“To address these issue, social media platforms have implemented various measures, including authentication systems for prominence users and algorithms designed to detect and remove suspicious accounts. However, this problem remains persistent, and the ongoing growth of new methods to identify and block fraudulent accounts remains a urgency for both researchers and platform operatives19.”

• cites [19] Alsubaei, F. S. Detection of inappropriate tweets linked to fake accounts on twitter. Appl. Sci. 13, 3013 (2023). sourcePartly supported

  The Introduction (page unavailable) says Twitter has made numerous attempts to enforce its policy against fake accounts, yet users continue to evade those efforts. The paper also develops a fake-account detector, but it does not establish that social media platforms have implemented authentication systems for prominent users or algorithms to detect and remove suspicious accounts. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

May combine two claims.

Social media networks' legitimacy may be threatened by fake accounts, damaging users' confidence and reputations.

“Social media networks’ legitimacy may be threatened by accounts, which would have a damaging effect on users’ confidence and reputations.”

• cites [12] Bokolo, B. G. & Liu, Q. Advanced algorithmic approaches for scam profile detection on instagram. Electronic 13, 1571 (2024). sourcePartly supported

  The Literature Review (section 2; page unavailable) describes fake profiles as a threat to user safety and says scams prey on user confidence. The Conclusions (section 6; page unavailable) link detecting fake profiles to a more trustworthy online community. These passages support the concern about confidence and trust, but do not address damage to users’ reputations. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim. May combine two claims.

Fake accounts contribute to spreading false information, illegal activities such as identity theft and phishing, and rumours, which may create social rigidities and divergences.

“There are several concerns such as spreading false information, engaging in illegal activities such as character burglary and phishing, and rumours, which may create rigidities and divergences in society7.”

• cites [7] Chakraborty, P., Shazan, M. M., Nahid, M., Ahmed, M. K. & Talukder, P. C. Fake profile detection using machine learning techniques. J. Comput. Commun. 10, 74–87 (2022). sourceSupported

May combine two claims.

The proposed model achieves an F1 score of 98%, recall of 98%, precision of 98.3%, and accuracy of 98.24% on Instagram fake profile detection.

“An F1 score of 98%, a recall of 98%, a precision of 98.3%, and an accuracy of 98.24% are all achieved by the new, perfectly accurate model that is used in the proposed research.”

May combine two claims.

The dataset used contains 200 fake Instagram accounts and 994 real accounts, with eight account authenticity features.

“The dataset consists of 200 fake accounts and 994 real accounts, respectively with eight account authenticity features.”

• AssumptionRests on an assumption: The open-source dataset's labels of real and fake Instagram accounts are accurate.
• AssumptionRests on an assumption: The 200 fake and 994 real Instagram accounts in the dataset are representative of Instagram fake profiles generally.

May combine two claims.

Random forest achieved accuracy 0.75, precision 0.77, recall 0.75, and F1-score 0.74.

“| Random forest | 0.75 | 0.77 | 0.75 | 0.74 |”

Probably not a substantial claim: may not pass S3 (load-bearing).

scale_pos_weight tuning selected scale_pos_weight=2.05 as the best value.

“| scale_pos_weight | scale_pos_weight | [1, 2, 3, 4] | 2.05 |”

May combine two claims.

Training loss starts high, decreases rapidly in the early epochs, and stabilizes at a low value after around 20 epochs.

“The training loss, shown by the blue line, starts high but reductions rapidly in the early epochs, indicating that the model is swiftly learning from the training data. As training progresses, the loss continues to decline and eventually stabilizes at a low value after around 20 epochs”

Probably not a substantial claim: may not pass S3 (load-bearing).

GridSearchCV selected max_depth=10 as the best value.

“| GridSearchCV | max_depth | [None, 10, 20] | 10 |”

May combine two claims.

Macro and weighted average F1-score are 0.98.

“The macro average and weighted average of the F1-score also stand at 0.98”

May combine two claims.

The confusion matrix shows balanced classification with minimal false positives (3.7%) and zero false negatives.

“A related confusion matrix shows balanced classification with minimal false positives (3.7%) and zero false negatives, which reflects reliability in identifying a real profile from a fake profile.”

• AssumptionRests on an assumption: The real/fake labels in the dataset are correct.

The quote only partly states this claim.

The proposed model achieved 98% accuracy in comparison with previous models.

“| Proposed | 98% |”

Probably not a substantial claim: may not pass S3 (load-bearing).

scale_pos_weight tuning selected learning_rate=0.05 as the best value.

“| scale_pos_weight | learning_rate | [0.01, 0.05, 0.1] | 0.05 |”

The quote only partly states this claim.

The proposed model's accuracy (98%) is higher than that of six prior models (93%, 94%, 93%, 96%, 83%, 94%).

“| S. D. Munoz et al62 | 93% |”

• Also stated in: § Result and discussion · plain

  “| G. Sonowal et al.63 | 94% |”

• Also stated in: § Result and discussion · plain

  “| Proposed | 98% |”

• cites S. D. Munoz et al62Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites G. Sonowal et al.63 Sonowal, G., Balaji, V. & Kumar, N. A model to detect fake profile on Instagram using rule-based approach. CSI Trans. ICT 1–11(12), 95 (2024). sourceNot checked — no full text of the source
• cites M. J. Ekosputra et al.64 Ekosputra, M. J., Susanto, A., Haryanto, F. & Suhartono, D. Supervised machine learning algorithms to detect instagram fake accounts. In 2021 4th International Seminar on Research of Information Technology and Intelligent Systems (ISRITI) 396–400 (IEEE, 2021). sourceNot checked — no full text of the source

• AssumptionRests on an assumption: The accuracies reported by prior studies are comparable to the proposed model's evaluation.

May combine two claims.

SVM achieved accuracy 0.70, precision 0.70, recall 0.70, and F1-score 0.70.

“| SVM | 0.70 | 0.70 | 0.70 | 0.70 |”

May combine two claims.

Macro and weighted average precision are both 0.98.

“In the case of precision values both from macro average and weighted average, their score stays constant at 0.98 in both aspects.”

May combine two claims.

XGBoost achieved accuracy 0.75, precision 0.78, recall 0.75, and F1-score 0.74.

“| XGBoost | 0.75 | 0.78 | 0.75 | 0.74 |”

May combine two claims.

Decision tree achieved accuracy 0.76, precision 0.78, recall 0.76, and F1-score 0.76.

“| Decision tree | 0.76 | 0.78 | 0.76 | 0.76 |”

The quote only partly states this claim.

Cross-validation analyses indicated that SMOTE boosted minority class representation.

“SMOTE boosted the minority class representation”

Probably not a substantial claim: may not pass S3 (load-bearing).

GridSearchCV selected min_samples_split=2 as the best value.

“| GridSearchCV | min_samples_split | [2, 5] | 2 |”

Probably not a substantial claim: may not pass S3 (load-bearing).

scale_pos_weight tuning selected n_estimators=100 as the best value.

“| scale_pos_weight | n_estimators | [50, 100, 150] | 100 |”

Probably not a substantial claim: may not pass S3 (load-bearing).

GridSearchCV selected n_estimators=100 as the best value for the Random Forest model.

“| GridSearchCV | n_estimators | [50, 100] | 100 |”

May combine two claims.

Training accuracy remains close to 100% from the beginning and stabilizes around this level over 40 epochs.

“The training accuracy, signified by the blue line, remains dependably high, close to 100%, from the beginning and stabilizes around this level.”

May combine two claims.

The proposed model achieved accuracy 0.98, precision 0.98, recall 0.97, and F1-score 0.98.

“| Proposed | 0.98 | 0.98 | 0.97 | 0.98 |”

• AssumptionRests on an assumption: The real/fake labels in the dataset are correct.

May combine two claims.

Logistic regression achieved accuracy 0.73, precision 0.74, recall 0.73, and F1-score 0.72.

“| Logistic regression | 0.73 | 0.74 | 0.73 | 0.72 |”

May combine two claims.

Validation loss starts elevated, decreases at a constant rate in earlier epochs, stabilizes after nearly 10 epochs, and remains nearly flat for subsequent epochs.

“Similarly, the validation loss, in orange, starts at an elevated value and continues lowering at a constant rate, as shown for the earlier epochs. Unlike the loss in training, the levels for loss in validation stabilize after nearly 10 epochs and goes nearly flat for the subsequent epochs of training.”

May combine two claims.

The approach uses XGBoost, SMOTE, and Random Forest, with scale_pos_weight and GridSearchCV optimization, resulting in high recall, precision, accuracy, and F1 score.

“It utilizes innovative machine learning techniques, including XGBoost, SMOTE and Random Forest, with scale_pos_weight and GridSearchCV optimization techniques resulting in high recall, precision, accuracy, and F1 score.”

May combine two claims.

The approach achieved overall Recall 98.0%, Precision 98.3%, accuracy 98.24%, and F1-Score 98.34%.

“Our approach has an overall Recall: 98.0%, Precision: 98.3%, accuracy of 98.24%, with F1-Score: 98.34% respectively demonstrating its ability to effectively distinguish between real and fake profiles.”

• AssumptionRests on an assumption: The real/fake labels in the dataset are correct.

Classified as probably not a substantial claim.

GridSearchCV-based hyperparameter tuning was conducted to fine-tune Random Forest, ensuring optimal depth and tree count to balance performance and generalization.

“First, GridSearchCV-based hyperparameter tuning was conducted to fine-tune Random Forest, ensuring optimal depth and tree count to balance performance and generalization.”

May combine two claims.

Fake accounts contribute to the spread of harmful content and misinformation, which can influence civic belief, generate conflicts, and erode trust in online civilizations.

“Fake accounts pay to the spread of harmful content, and misinformation, which can influence civic belief, generate conflicts, and erode trust in online civilizations.”

• cites [8] Khaled, S., El-Tazi, N. & Mokhtar, H. M. O. Detecting Fake Accounts on Social Media. In Proceedings - 2018 IEEE International Conference on Big Data, Big Data 2018 3672–3681 (Institute of Electrical and Electronics Engineers Inc., 2019). https://doi.org/10.1109/BigData.2018.8621913. sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

SMOTE was used to correct class imbalance, ensuring the model was exposed to a balanced number of real and fake accounts before the train-test split.

“In order to correct class imbalance within the dataset, SMOTE is used, which ensured the model is exposed to a balanced number of both real and fake accounts prior to dividing the dataset into preparation and justification sets.”

May combine two claims.

The pre-processed dataset was split into training and test sets at an 80:20 ratio, with class distributions preserved across subsets.

“Subsequently, the pre-processed dataset was split into preparation and test sets with 80:20, safeguarding that all class distributions are followed over subsets.”

• AssumptionRests on an assumption: The 80:20 train-test split preserves class distributions across subsets.

Probably not a substantial claim: may not pass S3 (load-bearing).

For Random Forest, GridSearchCV-based hyperparameter tuning with three-fold cross-validation was used to tune n_estimators, max_depth, and min_samples_split.

“For RF, GridSearchCV-based hyperparameter tuning with three-fold cross-validation to tune important parameters including the number of estimators tree depth (`max_depth`), (`n_estimators`), and the minimum number of samples needed to split a node (`min_samples_split`) are carried.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

Social media data analysis is central to proper detection and response to growing fake accounts.

“In general, social media data analysis stands at the heart of proper detection and response to growing fake accounts across the systems21.”

• cites [21] Sahoo, S. R. & Gupta, B. B. Hybrid approach for detection of malicious profiles in twitter. Comput. Electr. Eng. 76, 65–81 (2019). sourceNot checked — no full text of the source

May combine two claims.

The proposed method provides state-of-the-art methods for reducing the prevalence of false accounts, improving efficiency and trustworthiness of online identity verification systems.

“This paper provides state-of-the-art methods for reducing the prevalence of false accounts, which will improve the efficiency and trustworthiness of identity verification systems used online.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Cross-validation was used for strict evaluation to verify consistency.

“Feature engineering was performed in the dataset; the process also involved a strict evaluation by making use of cross-validation for verifying consistency.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Fake accounts are used to influence civic belief, change the work atmosphere, and destroy confidence in social networks.

“Fake accounts are used to influence civic belief, change the work atmosphere, and destroy confidence in social networks.”

• cites [14] Shahbazi, M. & Bunker, D. Social media trust: Fighting misinformation in the time of crisis. Int. J. Inf. Manage. 77, 102780 (2024). sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Fraudulent profiles create challenges for businesses that use social media for selling and appointment, as they may target fake or inactive accounts.

“Fraudulent profiles can create challenges for businesses that utilize social media for selling and appointment, as they may inadvertently target numerous fake or inactive accounts.”

• cites [18] Kaushik, K., Bhardwaj, A., Kumar, M., Gupta, S. K. & Gupta, A. A novel machine learning-based framework for detecting fake Instagram profiles. Concurr. Comput. Pract. Exp. 34, e7349 (2022). sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

SMOTE synthesized samples of the smaller class augmented its representation and improved the model's ability to classify fake accounts.

“SMOTE synthesized samples of the smaller class augmented the representation of this class and improved the model’s aptitude to classify fake accounts.”

• AssumptionRests on an assumption: SMOTE-generated synthetic samples for the minority class are representative and do not introduce harmful noise.

May combine two claims.

The presence of spam accounts erodes confidence in the online environment, leading users to question the credibility of information.

“The presence of spam accounts can lead to an eroding self-confidence in online environment, since the users are more likely to question the credibility of the information they encounter.”

• cites [11] Pasieka, N., Kulynych, M., Chupakhina, S., Romanyshyn, Y. & Pasieka, M. Harmful Effects of Fake Social Media Accounts and Learning Platforms. In CPITS 258–271 (2021).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Most detection methods apply machine learning algorithms on large amounts of data generated from social networking sites.

“Most of them apply the machine learning algorithms on huge amounts of data generated from such social sites20.”

• cites [20] Patel, K., Agrahari, S. & Srivastava, S. Survey on Fake Profile Detection on Social Sites by Using Machine Learning Algorithm. In ICRITO 2020 - IEEE 8th Int. Conf. Reliab. Infocom Technol. Optim. (Trends Futur. Dir. 1236–1240 (2020) https://doi.org/10.1109/ICRITO48877.2020.9197935. sourceNot checked — no full text of the source

Classified as probably not a substantial claim. May combine two claims.

Hyperparameter tuning and early stopping were performed for the Random Forest and XGBoost classifiers.

“As experimental work in proposed model fine-tuning included hyperparameter tuning and early stopping of the Random Forest as well as the XGBoost classifiers are performed.”

Classified as probably not a substantial claim. May combine two claims.

Fake-account activities can perform harmful illicit tasks including spoofs, phishing, or spamming, posing threats to users' privacy and safety.

“Fake-account activities can perform some harmful illicit tasks, including spoofs, phishing, or spamming, thereby potentially posing threats to the private and safe use of other users’ accounts in social networks.”

• cites [15] Choudhary, R. R., Paliwal, S. & Meena, G. Image forgery detection system using VGG16 UNET model. Proced. Comput. Sci. 235, 735–744 (2024). sourceNot checked — no full text of the source
• cites [16] Mughaid, A. et al. A novel machine learning and face recognition technique for fake accounts detection system on cyber social networks. Multimed. Tools Appl. 82, 26353–26378 (2023). sourceNot checked — no full text of the source

May combine two claims.

XGBoost was optimized using scale_pos_weight to handle class imbalance, preventing data duplication and reducing computational overhead.

“Second, rather than oversampling, XGBoost was optimized using scale_pos_weight to handle class imbalance effectively, which prevents data duplication and reduces computational overhead.”

Probably not a substantial claim: may not pass S3 (load-bearing).

False stories can be used to spread information and sway public opinion on important subjects including health information.

“False stories can be used to spread information and sway public opinion on a variety of subjects, including important concerns and health information13.”

• cites [13] Nedaković, A., Hasselgren, A., Kralevska, K. & Gligoroski, D. Hyperledger fabric platform for healthcare trust relations—Proof-of-concept. Blockchain Res. Appl. 4, 100156 (2023). sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

For XGBoost, scale_pos_weight was set to mitigate class imbalance, early stopping with early_stopping_rounds=10 was used to avoid overfitting, and logloss was the evaluation metric.

“For XGBoost, also the model is optimized by setting `scale_pos_weight` to mitigate class imbalance, enabling early stopping with `early_stopping_rounds = 10` to avoid overfitting, and using the evaluation metric of `logloss`.”

Classified as probably not a substantial claim.

Experiments based on Random Forest and XGBoost classifiers have proven to perform well in dealing with structured data.

“Experiments based on RF and XGBoost (XGB)55 classifiers have proven to perform well in dealing with structured data.”

• cites [55] Tang, Y., Zhang, D., Li, K.-C. & Yuan, J. Uncovering malicious accounts in online social networks using xgboost and graph convolution networks. In IEEE Trans. Consum. Electron. (2024). sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

SMOTE has been used for handling data imbalance, but resampling sometimes may introduce noise that degrades performance.

“SMOTE has been used for handling data imbalance, but resampling sometimes may introduce noise that degrades performance.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Steadily increasing accuracy across epochs means the model is learning effectively.

“Steadily increasing accuracy across epochs means that the model is learning effectively”

Classified as probably not a substantial claim.

All models were evaluated using recall, precision, accuracy, and F1-score to ensure fair comparisons.

“All models were evaluated using recall, precision, accuracy, and F1-score to ensure fair comparisons.”

May combine two claims.

Advanced models like BERT require substantial amounts of labeled textual data, significant computational assets, and extensive fine-tuning, which were beyond the practical scope of this investigation.

“Second, advanced models like BERT naturally require substantial amounts of labeled textual data, significant computational assets, and extensive fine-tuning, all of which were beyond the practical scope of this investigation.”

• cites [60] Kenton, J. D. M.-W. C. & Toutanova, L. K. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of naacL-HLT vol. 1 (Minneapolis, Minnesota, 2019).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The model was robust against any variation in dataset, implying generalization across various scenarios.

“Further analysis showed that the model was robust against any variation in dataset, thereby implying generalization across various scenarios.”

May combine two claims.

SMOTE addresses class imbalance, allowing the model to better identify fake profiles, as shown by better recall metrics.

“SMOTE addresses class imbalance, which allows the model to better identify fake profiles, as shown by better recall metrics.”

May combine two claims.

The hyperparameter adjustments collectively improved robustness, efficiency, and generalization of both models, leading to better classification performance.

“These adjustments collectively improved the robustness, efficiency, and generalization capability of both models, leading to better classification performance.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Many earlier fake profile detection approaches focused narrowly on particular features or relied on outdated methods, making them inefficient against changing tactics of fake account creators.

“The literature review points out the fact that many earlier approaches focused narrowly on particular features or relied upon outdated methods that make them inefficient against the changing tactics used by the people who are creating these fake accounts.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

These results affirm the model's applicability in real-world scenarios and provide a basis for future research.

“These results affirm the model’s applicability in real-world scenarios and deliver a basis for future research”

• AssumptionRests on an assumption: The real/fake labels in the dataset are correct.

Probably not a substantial claim: may not pass S4 (based).

The literature rarely explores optimized hybrid frameworks that combine multiple models with fine-tuned parameters for improved classification accuracy.

“The literature rarely explores optimized hybrid frameworks that combine multiple models with fine-tuned parameters for improved classification accuracy.”

May combine two claims.

The study kept evaluation metrics, pre-processing step, and dataset partition the same between models to achieve robust and clear performance comparisons, increasing the validity of the proposed model's supremacy.

“The study keeps evaluation metrics, the pre-processing step and dataset partition all the same between models in a way to achieve robust, as well as clear performance comparisons; this increases the validity of proposed model’s supremacy.”

May combine two claims.

The hybrid model's total complexity incorporates both Random Forest and XGBoost and ensures optimal performance while balancing training efficiency.

“The hybrid model’s total complexity, incorporating both approaches, is , ensuring optimal performance while balancing training efficiency.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

While SMOTE has been used for class balancing, improper oversampling can introduce noise and hinder model generalization.

“Additionally, while SMOTE has been used for class balancing, improper oversampling can introduce noise and hinder model generalization.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Hyperparameter optimization techniques such as GridSearchCV have not been applied to most previous studies, resulting in suboptimal model configurations.

“Moreover, hyperparameter optimization techniques GridSearchCV have not been applied to most studies, which again results in suboptimal configurations of the model.”

Probably not a substantial claim: may not pass S3 (load-bearing).

SMOTE has been extensively used to handle class imbalance in fake account detection tasks to improve model generalizability.

“Likewise, SMOTE has been extensively used to handle the class imbalance issue in fake account detection tasks to improve model generalizability56.”

• cites [56] Shah, A., Varshney, S. & Mehrotra, M. Detection of fake profiles on online social network platforms: Performance evaluation of artificial intelligence techniques. SN Comput. Sci. 5, 489 (2024). sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The paper integrates machine learning techniques with extensive feature engineering and evaluation, presenting a scalable and effective solution to improving online trust.

“This paper integrates the finest cut of machine learning techniques with extensive feature engineering and evaluation and, thus, presents a scalable and effective solution to improving online trust.”

Classified as probably not a substantial claim. The quote only partly states this claim.

All baseline models use SMOTE for class balancing.

“All these above-mentioned models use SMOTE for class balancing.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

The primary gap in fake profile detection research lies in limitations of existing methods, which often struggle with class imbalance, ineffective feature utilization, and suboptimal performance due to lack of hyperparameter tuning.

“The primary gap in fake profile detection research lies in the limitations of existing methods, which often struggle with class imbalance, ineffective feature utilization, and suboptimal performance due to a lack of hyperparameter tuning.”

May combine two claims.

The model is overfitting, likely memorizing training data rather than generalizing to unseen data, because validation accuracy does not improve despite high training accuracy.

“The fact that the justification accuracy does not advance much, despite the high training correctness, indicates that the model is overfitting. It is likely memorizing the patterns in the training data rather than learning to generalize to unseen-data, which is why the validation performance stays relatively static.”

May combine two claims.

XGBoost is more computationally expensive at training time but efficient at prediction time.

“Thus, XGBoost is more computationally expensive at training time but efficient at prediction time.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Most conventional approaches have focused on manually designed features including user behaviour patterns, language clues, and social network topology, but struggle to handle intelligent fake profiles that mimic human interactions.

“Most conventional approaches have focused on manually designed features including user behaviour patterns, language clues, and social network topology. The drawback to these methods, however, is that they struggle to handle intelligent fake profiles with the ability to mimic human interactions.”

May combine two claims.

Through parallelization and early stopping, XGBoost's runtime is reduced, making the ensemble model feasible for large-scale fake profile detection.

“Through parallelization and early stopping, XGBoost’s runtime is reduced, making the ensemble model feasible for large-scale fake profile detection.”

May combine two claims.

Compared to baseline models Logistic-Regression, SVM, and DT, the proposed approach validates superior performance in recall, accuracy, F1-scores, and precision.

“Equated to baseline models Logistic-Regression, SVM, and DT, our approach validates superior concert in recall, accuracy, F1-scores, and precision underscoring its usefulness.”

May combine two claims.

The proposed framework solves the problem of class imbalance and displays outstanding adaptability in detecting dynamic behavioural patterns in fake accounts.

“Our proposed framework solves the problem of class imbalance and displays outstanding adaptability in the detection of dynamic behavioural patterns in fake accounts.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Traditional ML models Random Forest and XGBoost have been predominantly used in previous fake profile detection literature but are known to be inefficient for class imbalance and feature selection.

“Since traditional ML models RF and XGBoost have been predominantly used in previous literature on fake profile detection; however, they are known to be inefficient for class imbalance and feature selection.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Hybrid approaches combining various classifiers have been introduced to further enhance detection efficiency.

“Also, hybrid approaches based on combining various classifiers have been introduced to further enhance detection efficiency.”

Probably not a substantial claim: may not pass S4 (based).

Many prior studies rely on single-model approaches, such as Random Forest or XGBoost, without leveraging their complementary strengths.

“Many prior studies rely on single-model approaches, such as Random Forest or XGBoost, without leveraging their complementary strengths.”

Probably not a substantial claim: may not pass S3 (load-bearing).

XGBoost has exhibited enhanced classification performance due to its capability of capturing complex feature interactions.

“work that leveraged XGBoost has exhibited enhanced classification performance as a result of the capability of capturing complex feature interactions.”