Online Social Networks involve a huge amount of people from all over the world and it has become a big part of their life. People use social networks to share their feelings, to make new friends, to set up new businesses, to connect with friends and family and what not. The Online Social Networks provides a great advantage to individuals in different ways but it also suffers with some disadvantages. There are many people who use these networks to cause harm to others by making fake accounts on these networks. For detection of such fake and genuine accounts we can use machine learning algorithms. The machine learning algorithms are applied for the prediction and classification of datasets through the different models that are prepared. It sometimes become difficult to differentiate between the results of different models and so we to use a hybrid approach of machine learning algorithm can make this task easy. In our work we compared the 8 different combinations of classification algorithms and calculated their accuracy on the dataset of an Online Social Network. We used the combination of Random Forest, Support Vector Machine, Logistic Regression, KNN, and Decision Trees. After comparing the result of each hybrid approach, we concluded that the best accuracy was obtained by combination of SVM and Logistic Regression and Neural Network. So, we proposed a model for the detection of fake account with the hybrid approach giving the best accuracy among all the combinations.
Machine Learning is a branch of artificial intelligence (AI) which is able to provide a system the ability to act without being programmed explicitly. It is used in many fields like Google cars, recommendation engines, friend suggestions in social media networks, shopping apps, cybercrimes etc.
Machine Learning has made a phenomenal change in the way how data was extracted and interpreted by replacing the old statistical techniques. Classifications of machine learning techniques are: Reinforcement, Supervised and Unsupervised Machine Learning.
Our work is concerned with the Classification algorithms that come under the Supervised Machine Learning. Classification is a supervised learning approach in which the machine takes the input data learns from that data and then further classifies the testing data according to its training data.
Although classification algorithms (Support Vector Machine, Logistic Regression, Decision Tree, Random Forest, Artificial Neural Network) can be used separately and individually but in our system we are developing a hybrid model combining two or three machine learning models has helped in increasing the accuracy of the model and its predicative power. The fact that which hybrid model will perform better is unknown, but it is also affected by the dataset provided and also the feature selection. The concept to develop a hybrid model is in a two- stage manner, first using clustering or classification techniques for pre-processing of data and in second stage the output of the first stage to build second stage predictive classifier. It can be made using different algorithms of supervised or unsupervised learning but in our work, we developed the model using classification algorithms of supervised learning. Our main contribution is to propose a hybrid approach of machine learning algorithms and to compare the hybrid of different classification algorithms. Eight different experiments were conducted, and the accuracy thus obtained was compared.
The total number of users in online social networking sites is continuously increasing and with that the number of fake accounts is also increasing. As in September 2019, monthly active users on Facebook are 2.45 billion worldwide. According to Alexa, after Google and YouTube the third most visited website is Facebook. In a survey it is found that there are a greater number of female accounts in the world than the total population of female. From this, we can infer how many fake profiles have been created. According to Statistics April 2018 stats report, Facebook has more than 336 million active Twitter accounts, but Facebook is the leader with 2,196 million users worldwide. In September 2019, monthly active users on Facebook are 2.45 billion, of which India has the most. 270 million users. People who log on to Facebook daily are approximately 1.62 billion. And among these 83 million accounts are fake on Facebook. This statistics was given by Facebook in their Wall Street reports (SOURCE: Zephoria Digital Marketing). Figure 1 shows the monthly active users in the year 2019 on various OSNs.
As the number of people using OSN increases, so does the fake social media accounts creation. The main motivational factor in identifying those fake accounts is the cyber-crime rate, as these accounts were created primarily to commit cyber robbery or to commit cybercrime anonymously or unidentified is a significant increase from last few years. Fake account owners also try to take advantage of people's kindness by composing fake messages and spreading false news through these fake accounts in order to steal money from sinless people. In addition, people want to create multiple accounts that don't belong to anyone, created just to raise votes in an online voting system, and receive referral incentives, as in online games.
The detection of fake accounts in OSN attracts many researchers, so several algorithms for detection of fake accounts have been developed using machine learning techniques and various functions to connect to the account. Spammers can also find ways to support such techniques. These security technologies provide sophisticated detection mechanisms that require the continuous development of new approaches to spam detection. The main hazards in detection of fake accounts are to achieve accuracy and response time in the analysis of characteristics. 1.2.
Modeling a Fake Profile Detection System is an old problem but due to the many challenges this problem presents there still exist a lot of gaps that have been identified and need to be worked upon. The many challenges this system presents have been listed below:
The data is not readily available: accounts on online social networks are highly private and protected, so the networking sites do not reveal any account information to maintain the confidential nature and keep the trust of their users.
There is a lot of overlapping between genuine and fake accounts: At times the feature set of legitimate and fake accounts overlap, and this poses a considerable setback when it comes to training the neural network by making it learn the pattern to differentiate between them.
The number of parameters to process: The enormous number of parameters between learning and decision making is a major obstacle in developing systems for detecting fake accounts.
Selection of optimal features (variables) is a big challenge: When it comes to optimal feature selection, it needs to be really dealt with care as the performance of whole system depends on which features it’s taking into consideration for classification of fake and genuine accounts. And at times it’s really perplexing to decide on these optimal features.
Ability to handle noise in the data: Noise means missing or incorrect data which poses challenges while processing the dataset. There is no means by which we can make up for this lost information as such systems aren’t partition tolerant, so this adversely affects the outcome.
Heterogeneity in features. Single user multiple accounts.
Many of the times it resembles a legitimate transaction: At times the fake account activities are stacked up in close resemblance with the legitimate ones. Hence, it becomes difficult to comprehend them and abort them before they make it to completion. 1.3.
We can extend the evaluation of propose feature by testing on different social networking sites like Facebook, Twitter etc. as most the previous researches were done on any one social site among Facebook, Twitter, LinkedIn, Myspace etc.
The existing system does not work for the real time accurately on changing the features.
Identification of rumor sources on social media by using the content-based features. 1.4.
Fake Profile Detection is an old problem and there has been a lot of work done in providing an optimal solution. But the fact that the mannerism of fake accounts keeps on evolving with time and there are enormous numbers of challenges and gaps still left to tackle, this problem still has a lot of significance. In order to study the work already done on Fake Account Detection we searched articles and research papers on two major sources: i) general online indexing websites, ii) publisher databases. Examples of former are Research Gate, Towards Data Science, IEEE Explorer and Google Scholar and examples of latter are Scopus, Springer, ACM Digital Library and Elsevier databases.
The major machine learning techniques we used in detection of fake accounts are Neural Network, Support Vector Machine, Random Forest, and Hybrid Models for Comparative analysis of Fake Account Detection.
Yang et al. trained SVM using the ground- truth obtained by Ren Ren for detecting fake accounts. By making use of simple features like frequency of friend requests, accepted requests and peraccount clustering coefficient they trained the classifier and got 99% true-positive rate (TPR) and 0.7% false- positive rate (FPR). Íntegro draws out low-cost features from user-level activities to train the classifier for the identification of undetermined victims in social graph and used feature-based detection.
A different approach for hybrid was introduced by Mateen et al., by using content- based features like total number of tweets, hash tag ratio, URL’s ratio and some graph- based features also and used the dataset of Twitter. They also made a comparison J48, Decorate and Naïve Bayes in which Decorate was the best performer. Somya et al.’s approach was quite different from others for detection as they tried to detect the account as fake on the user’s homepage using Chrome extension which runs on the user site. Along with this they used Petri net based solution for the identification of source of malicious content running on Pn2 simulator environment.
Using a support vector machine and a
neural network, Khaled et al [
In our work we have implemented various algorithm to find the most efficient algorithm. To do so we have conducted several experiments and compared their results. Further, in this paper we have three sections which are briefly define below:
This section is followed by Section 2, System Architecture. In this section, flow diagram and architecture of our work is introduced and is described in brief.
In Section 3, Experimental Results, of up to now what modules we have implemented is shown along with pseudocode and discussed the various results produced by our system and have shown the outputs generated on various inputs in the form of the graph for the better understanding and algorithm of the technique is also mentioned in this section.
In Section 4, Conclusion, we provide an understanding of the overall conclusion of the proposed solution i.e. the combination of the techniques which is efficient than others and is given better accuracy.
Although fake profile detection is a robust field, but it has many challenges and gaps which we have discussed and have based our work on. There are a lot of existing solutions to fake profile detection but all of them have some or the other drawback. There is a lot of work already done in this field and a lot more needs to be done like improving upon the response time, prevention from fake accounts instead of detecting and dealing with their aftermaths. Our work is aiming to deliver a system which will have the highest accuracy and hence will be effective in prevention from such fake profiles by implementing and comparing different algorithms. This is done by ensemble machine learning technique which speeds up the training of neural networks and helps them to take decisions faster. Efficient parameter selection is also one of the major objectives of this work for which we are selecting six features manually which will give a better control on the output of neural networks. The proposed solution makes use of the hybrid of the machine learning techniques and combines their advantages and uses one to cancel out the loopholes of the other and hence delivering an efficient and cost-effective system.
In our proposed system we are aiming to design a hybrid system using artificial neural network, support vector machine and logistic regression that will be able to precisely and accurately detect fake profiles in online social network. Goal of the work is to maximize the accuracy and to minimize the time required by using hybrid approach of the Neural Network, Support vector machine and Logistic Regression.
Figure 2 depicts the flowchart of our system. The dataset which we have is partitioned into two sets, Train Dataset and Test Dataset in the ratio 4:1.The train dataset then goes into Support Vector Machine and Logistic Regression Classifier where classes are predicted. Then these classifiers are appended to a voting classifier where final decision of class is made. The output from voting classifier i.e. train data and the predicted class from voting classifier is fed to Neural Network classifier as input. After training has been completed, we get a Trained System on which Test dataset is ran to find the accuracy of the system.
No proposal can be modeled into a system without some experiments to support it. In this section we have included the results and outputs produced during experiment with our system and by our system under various inputs and parameters.
Each phase of our proposed system is briefly described in this section along with description, results at each stage are also provided.
For the model to work upon, there is a need for data collection. The dataset can be collected from various online platforms and can also be created by using Crawler. We have collected two datasets through online from well-known websites Kaggle and GitHub. But we worked on the dataset which is collected by Kaggle and in that we are using two CSV files corresponding to fake and genuine users. Figure 5 shows the sample of csv file. And the code for reading both the files are: genuineusers=pd.read_csv("users.csv") fakeusers= pd.read_csv(“fusers.csv") ,
Data pre-processing is used to achieve the better result from any machine learning model and data processing is used to clean the data from raw data we import the useful libraries which will rescale or clean our data and the libraries we import are numpy, panda, scikit-learn and from sklearn we import preprocessing to clean our data.
Now in the next part for data preprocessing we use feature extraction technique first we try the principal component analysis technique and then we use the genetic algorithm and then after we select the features manually and we compare the result obtained from three ways and we get better result from the manually selection of features and the features we select manually are: statuses_count followers_count friends_count favourites_count listed_count lang_code
The language code feature is of string type we convert it into integer. After calling extract feature function it prints the extracted feature name and describes the entire extracted feature in summarized by printing mean, quartile, count, std, min, max etc.
Figure 4 shows the data distribution in each column or feature in terms of count, mean, standard deviation, minimum and maximum values, and average of 25%, 50% and 75% of the data points when taken in ascending order.
As we are using the hybrid approach of the techniques in our proposed system, so we have done experiments with six techniques i.e. SVM, RF, LR, DTC, NN, KNN and finalize the techniques that gives the best result and they are Support Vector Machine, Logistic Regression and Neural Network. First we train our data using support vector machine independently and then we train our data on Logistic Regression independently and after analyzing the result of both the classification techniques we merge both the techniques to check the accuracy of both of them together and hybrid approach of both the techniques gives us the best result and after training the data from both the voting classifier is used to get the best result from both and then passing value for any one of them and then we use 5 fold cross validation technique to avoid the situation of overfitting as in k-fold cross validation technique dataset in divided into k folds where 1 fold is used for validation or testing while others are used for training and in these way we can avoid the situation of overfitting. After getting the score of each fold final estimated score is printed and in these we got 0.91 and the accuracy on testing dataset is 99.56.and after that the confusion matrix is plotted which will gives us the 261 true positive value and 7 false negative value and 29 false positive and 267 true negative value and then we plot the normalized confusion matrix which gives us all the four (TP,TN,FP,FN) values in percentage form along with precision, recall, f1 score and support and all these are evaluation criteria. For fake recall we got is 0.98 and for genuine it is 1.00 and f1 score for both is 0.99 and overall accuracy is0.99.
called epoch. For this instance, we have taken our epoch to be 10, total number of layers to be 3, it took approximately minutes and seconds to train the system with final accuracy and loss value to be respectively.
Now the output produced by several hybrid techniques. We have collected two datasets say, D1 and D2 and the difference between these datasets is in their size, D2 is large as compared to D1. D2 contains approx. 3500 rows while D1 contains approx. 1500 rows. The results that we have obtained with different algorithms on both datasets are different and D2 gives less system with less accuracy as compared to D1.
As we can see there is an accuracy difference between both datasets used by different algorithms so further, we will be working and showing results for only dataset, D1. We are using two csv files one is of genuine users and other one is of fake users.
Figure 7 shows the accuracy of each of our experimental model in ascending order and the model with highest accuracy being our trained system. Table 2 shows the results of the seven experiments that we performed using different combination of classification algorithms like Support Vector Machine, Random Forest, Logistic Regression, KNN with Neural Network. In the above table we can see that SVM, Log Reg, and NN is giving the maximum of true positive true negative resulting in maximum accuracy of all.
If we look at the system designs, majority of implementations for fake account detection is either graph-based or featurebased and they may use the graph analysis techniques or machine learning techniques to identification of accounts as fake or real. In our proposed framework we use feature-based dataset and selected the features manually. This approach is based upon the user-level activities and the user’s account details. We are comparing the hybrid approach of different classification algorithms and pass them in voting classifier and then pass the result in Neural network what we got from the voting classifier. In addition to our satisfying conclusion, we have maintained the highest accuracy in detecting fake accounts by testing and training the dataset on different hybrid approach of classification algorithms. The results show the increase of the accuracy results of the different classification algorithm.
Hoda MO Mokhtar. "Detecting fake accounts on social media." 2018 IEEE International Conference on Big Data (Big Data). IEEE, 2018. [21] Gupta, Aditi, and Rishabh Kaushal. "Towards detecting fake user accounts in facebook." 2017 ISEA Asia Security and Privacy (ISEASP). IEEE, 2017. [22] Benevenuto, Fabricio, et al. "Detecting spammers on twitter." Collaboration, electronic messaging, anti-abuse and spam conference (CEAS). Vol. 6. No. 2010. 2010. [23] Stein, Tao, Erdong Chen, and Karan Mangla. "Facebook immune system." Proceedings of the 4th workshop on social network systems. 2011.