Click Fraud Detection Research Articles

A transfer learning framework towards identifying behavioral changes of fraudulent publishers in pay-per-click model of online advertising for click fraud detection

The absence of a publicly available user-click dataset makes the task of fraudster identification particularly challenging to detect click fraud in online advertising. However, the task becomes more complicated with concept drift, where a publisher appears in distinct sets with the same pattern but differs in real status labels. Fraud-detection algorithms developed the actual status labels. The reliability of the predictions made by learning models needs to be examined to deal with the concept drift concerning the changing behaviour of fraudulent publishers without re-training the predictive model from scratch. However, the scarcity of pre-trained models aggravates the issue. Thus, we proposed a 1D-convolutional neural network-based fraudsters identification network (FINet) that deals with such challenges using a model trained on a correlated prediction modelling problem using transfer learning. To overcome the scarcity of pre-trained models, we designed a model FINet trained on the correlated dataset TalkingData Ad Tracking Fraud Detection (TDA) and utilized the weights of the trained model to learn a model on the FDMA2012 dataset better. This predicted modelling on a distinct but related problem is re-used for accelerating the training and improving the FINet’s fraudster identification performance. The proposed FINet’s behaviour is investigated based on eight optimizer algorithms in terms of Average Precision, Recall, F1-score, and AUC scores and performance generalization is verified by conducting experiments with existing state-of-art deep learning models. The implementation results are notably higher than the existing state-of-the-art deep learning models and exhibit effective performance towards fraudster’s identification in detecting click fraud.

Stacked Generalization Architecture for Predicting Publisher Behaviour from Highly Imbalanced User-Click Data Set for Click Fraud Detection

Stacked Generalization Architecture for Predicting Publisher Behaviour from Highly Imbalanced User-Click Data Set for Click Fraud Detection

Click fraud detection for online advertising using machine learning

Advertising corporations have moved their focus to online and in-App advertisements in response to the expansion of digital technologies and social media. Online advertising represents the primary revenue source for advertising networks that serve as the middlemen between advertisers and advertisement publishers. The advertising networks pay the publisher of the advertisement based on the number of clicks through to advertisers, following the pay-per-click (PPC) payment scheme. However, there is a growing security issue with this payment approach known as Click Fraud. Click fraud is the illegal process of clicking on pay-per-click advertisements to increase publishers' revenue or deplete advertisers' budgets. Artificial intelligence techniques have been increasingly employed to solve complicated challenges in different research areas, including cybersecurity, to achieve unexpected outcomes. In this paper, several Machine Learning models were constructed to establish whether the user is a human or a bot and conducted a comparative performance analysis using a set of evaluation metrics. We used a real captured dataset detailing Internet users' behavior while navigating websites. We extracted a set of features related to users’ behaviors, including the number of webpages viewed during the browsing session, the duration of the browsing journey, and the actions performed. The empirical results revealed that all the considered models obtained good results, where the random forest algorithm surpassed all other algorithms in all evaluation metrics.

AI-Based Techniques for Ad Click Fraud Detection and Prevention: Review and Research Directions

Online advertising is a marketing approach that uses numerous online channels to target potential customers for businesses, brands, and organizations. One of the most serious threats in today’s marketing industry is the widespread attack known as click fraud. Traffic statistics for online advertisements are artificially inflated in click fraud. Typical pay-per-click advertisements charge a fee for each click, assuming that a potential customer was drawn to the ad. Click fraud attackers create the illusion that a significant number of possible customers have clicked on an advertiser’s link by an automated script, a computer program, or a human. Nevertheless, advertisers are unlikely to profit from these clicks. Fraudulent clicks may be involved to boost the revenues of an ad hosting site or to spoil an advertiser’s budget. Several notable attempts to detect and prevent this form of fraud have been undertaken. This study examined all methods developed and published in the previous 10 years that primarily used artificial intelligence (AI), including machine learning (ML) and deep learning (DL), for the detection and prevention of click fraud. Features that served as input to train models for classifying ad clicks as benign or fraudulent, as well as those that were deemed obvious and with critical evidence of click fraud, were identified, and investigated. Corresponding insights and recommendations regarding click fraud detection using AI approaches were provided.

A hybrid data‐level sampling approach in learning from skewed user‐click data for click fraud detection in online advertising

AbstractOne of the challenging issues in user‐click data of online advertising is the uneven class distribution which biases classification models. Resampling the data is a popular choice for obtaining class balance. However, oversampling results in overfitting, whilst under‐sampling results in information loss. Moreover, enhancing separability between samples, where the classes overlap closer to the decision boundary, is another challenge, which requires a careful pruning of instances towards increasing the separability in data space. Therefore, in this work, a new hybrid data sampling algorithm SMOTEOSS is designed and evaluated, concatenating the synthetic minority oversampling technique (SMOTE) followed by one‐sided selection (OSS) to balance the class distribution. The working of SMOTEOSS is twofold‐ first, it oversamples the under‐represented class distribution using the SMOTE by generating synthetic instances. However, the generation of synthetic instances closer to the decision boundary directly influences the learning model's decision‐making. Utilising OSS, the proposed method then identifies TOMEKLINKS and eliminates the noisy majority instances whilst eliminating the redundant instances. The proposed method's effectiveness is validated on the FDMA 2012 dataset against 10 state‐of‐the‐art sampling methods utilising the gradient tree boosting learning model. To authenticate SMOTEOSS, a fair comparison is made by conducting experiments on other 10 benchmark imbalanced datasets using 10‐fold cross‐validation. Performance is measured using average precision, recall, F1‐score, G‐mean, the area under curve (AUC) and reduction rate. Results showed that the designed hybrid methodology is an efficient alternative to existing sampling methods. The Wilcoxon signed‐rank test is employed to demonstrate significant differences amidst the proposed and conventional sampling algorithms.

Click Fraud Detection of Online Advertising–LSH Based Tensor Recovery Mechanism

Online advertising utilizes the Internet technique to deliver marketing messages to promotional consumers. It is growing in recent years to facilitate the increasing demands of electronic commerce. Advertisers bid and pay for the advertisement whenever potential customers click it. The way of Pay-Per-Click (PPC) is vulnerable to malicious clicks that mimic real user behaviour to trick the platform into counting their clicks as legitimate. It causes massive financial losses on advertisers and also significantly reduces the credibility of online advertising platforms. The common strategies to detect fraud clicks are dynamically tailoring and interpreting data based on the machine learning model. These algorithms treat multi-dimensional data as an individual feature vector or matrix, making it different to explore intrinsic relations among a sequence of data. Million daily fraud clicks on various types further disperse the focus of models and result in relatively low efficiency for the current fraud prediction system. To tackle the fraud click problem, we introduce a tensor-based mechanism to predict fraud clicks. This paper considered reconstructing data into a high-rank tensor, implement tensor decomposition and transformation to explore hidden information under each data and explore the joined effect among a sequence of data. The proposed tensor transformation algorithm with locality-sensitive hashing (LSH) is tested by extensive experiments using real-world data. Compared with the state-of-art machine learning algorithms, our model can achieve significant performance in terms of accuracy and prediction-recall rate.

Feature space transformation of user-clicks and deep transfer learning framework for fraudulent publisher detection in online advertising

The concept drift is a challenge in click fraud detection wherein frequent changes in the actual status label of publishers complicate the identification of publishers’ fraudulent behavior. However, using transfer learning by leveraging the knowledge from previously learned domains to newer domains can make these differences more accessible while saving training time and improving the model’s performance. But the absence of other user-click datasets available publicly poses complexity in using transfer learning. Therefore, to use transfer learning towards predicting the publisher’s conduct concerning change in their labels, this work aims to transform 1D user-click non-image features into a 2D graphical image. The work proposes a deep convolution neural network-based transfer learning (DCNNTr) framework that utilizes different pre-trained Deep Convolutional Neural Network (DCNN) models as powerful feature extractors that leverage prior learnings to avert learning from scratch. The robust features extracted by the feature extractors help identify the conduct of publishers and classify them as fraudulent or non-fraudulent from 2D graphical images using machine learning models. By leveraging the weighted layers in extracting features, DCNN models utilize their special properties of being computationally efficient and locally focused. We evaluated the designed model on the FDMA2012 user-click dataset using precision, recall, F1-score, and AUC. Current work uniquely transforms the time series user-click non-image data into an image form. The experimental results demonstrate that features extracted with DenseNet121 followed by GTB have identified the fraudulent publishers with an average precision score of 79.8%.

BCFDPS: A Blockchain-Based Click Fraud Detection and Prevention Scheme for Online Advertising

Online advertising, which depends on consumers’ click, creates revenue for media sites, publishers, and advertisers. However, click fraud by criminals, i.e., the ad is clicked either by malicious machines or hiring people, threatens this advertising system. To solve the problem, many schemes are proposed which are mainly based on machine learning or statistical analysis. Although these schemes mitigate the problem of click fraud, several problems still exist. For example, some fraudulent clicks are still in the wild since their schemes only discover the fraudulent clicks with a probability approaching but not 100%. Also, the process of detecting a click fraud is executed by a single publisher, which makes a chance for the publisher to obtain illegal income by deceiving advertisers and media sites. Besides, the identity privacy of consumers is also exposed because the schemes deal with the plain text of consumers’ real identity. Therefore, in this paper, a blockchain-based click fraud detection and prevention scheme (BCFDPS) for online advertising is proposed to deal with the above problems. Specifically, the BCFDPS mainly introduces bilinear pairing to implicitly verify whether a consumer’s real digital identity is contained in a click message to significantly avoid click fraud and employs a consortium blockchain to ensure the transparency of the detection and prevention process. In our scheme, the clicks by machines or fraud ones by a human can be accurately detected and prevented by media sites, publishers, and advertisers. Furthermore, ciphertext-policy attribute-based encryption is adopted to protect the identity privacy of consumers. The implementation and evaluation results show that compared with the existing click fraud detection and prevention schemes based on machine learning and statistical analysis, BCFDPS achieves detection of each fraudulent click with a probability of 100% and consumes lower computation cost; furthermore, BCFDPS adds functions of consumers’ privacy protection and click fraud detection and prevention, compared to the existing blockchain-based online advertising scheme, by introducing limited communication cost ( 4,984 bytes) at lower storage cost.

An Ensemble Architecture Based on Deep Learning Model for Click Fraud Detection in Pay-Per-Click Advertisement Campaign

With the rapid development of online advertising, click fraud is a serious issue for the internet market. Click fraud is a dishonest attempt to improve a website’s profit or deplete an advertiser’s budget by clicking on pay-per-click advertisements. For an extended period, this illegal act has a threat to the industrial sectors. As a result, these businesses hesitate to advertise their items on mobile apps and websites, as numerous groups attempt to take advantage of themes. To safely advertise their services and products online, a robust mechanism is needed for efficient click fraud detection. To tackle this issue, an ensemble architecture of machine learning and deep learning is proposed to detect click fraud in online advertisement campaigns. The proposed ensemble architecture consists of a Convolutional Neural Network (CNN), and a Bidirectional Long Short-Term Memory network (BiLSTM) is used to extract hidden features, while the Random Forest (RF) is used for classification. The main objective of the proposed research study is to develop a hybrid DL model for automatic feature extraction from clicks data and then process through an RF classifier into two classes, such as fraudulent and non-fraudulent clicks. Furthermore, a preprocessing module is developed to preprocess data by dealing with categorical attributes and imbalanced data to enhance the reliability and consistency of the clicks data. In addition, different evaluation criteria are used to evaluate and compare the performance of the proposed CNN-BiLSTM-RF with the ensemble and standalone models. The experimental results indicate that our ensemble architecture achieved the accuracy of 99.19 ± 0.08%, precision 99.89 ± 0.03%, sensitivity 98.50 ± 0.11%, F1-score 99.19 ± 0.08% and specificity 99.89 ± 0.03%. Furthermore, our proposed architecture produced superior results compared to other developed ensemble and conventional models. Moreover, our proposed ensemble architecture can be used as a safeguard against click fraud for pay-per-click advertising to facilitate industries for the safe and reliable promotion of their products.

Click Fraud Detection Approaches to analyze the Ad Clicks Performed by Malicious Code

Mobile PR is an important component of the mobile app ecosystem. A major threat to this ecosystem’s long-term health is click fraud, which involves clicking on ads while infected with malware or using an automated bot to do it for you. The methods used to identify click fraud now focus on looking at server requests. Although these methods have the potential to produce huge numbers of false negatives, they may easily be avoided if clicks are hidden behind proxies or distributed globally. AdSherlock is a customer-side (inside the app) efficient and deployable click fraud detection system for mobile applications that we provide in this work. AdSherlock separates the computationally expensive click request identification procedures into an offline and online approach. AdSherlock uses URL (Uniform Resource Locator) tokenization in the Offline phase to create accurate and probabilistic patterns. These models are used to identify click requests online, and an ad request tree model is used to detect click fraud after that. In order to develop and evaluate the AdSherlock prototype, we utilise actual applications. It injects the online detector directly into an executable software package using binary instrumentation technology (BIT). The findings show that AdSherlock outperforms current state-of-the-art methods for detecting click fraud with little false positives. Advertisement requests identification, mobile advertising fraud detection are some of the keywords used in this article.

Quad division prototype selection-based k-nearest neighbor classifier for click fraud detection from highly skewed user click dataset

In online advertising, the user-clicks dataset based fraudulent publishers’ classification models exhibit poor performance due to high skewness in class distribution of the publishers. The nearest-neighbor based classification techniques are popularly used to reduce the impact of class skewness on performance. The Nearest-Neighbor techniques use Prototype Selection (PS) methods to select promising samples before classifying them for reducing the size of training data. Although Nearest-Neighbor techniques are simple to use and reduce the negative impact of the loss of potential information, they suffer from higher storage requirements and slower classification speed when applied on datasets with skewed class distributions. In this paper, we propose a Quad Division Prototype Selection-based k-Nearest Neighbor classifier (QDPSKNN) by introducing quad division method for handling uneven class distribution. The quad-division divides the data into four quartiles (groups) and performs controlled under-sampling for balancing class distribution. It reduces the size of the training dataset by selecting only the relevant prototypes in the form of nearest-neighbors. The performance of QDPSKNN is evaluated on Fraud Detection in Mobile Advertising (FDMA) user-click dataset and fifteen other benchmark imbalanced datasets to test its generalizing behaviour. The performance is also compared with one baseline model (k-NN) and four other prototype selection methods such as NearMiss-1, NearMiss-2, NearMiss-3, and Condensed Nearest-Neighbor. The results show improved classification performance with QDPSKNN in terms of precision, recall, f-measure, g-mean, reduction rate and execution time, compared to existing prototype selection methods in the classification of fraudulent publishers as well as on other benchmark imbalanced datasets. Wilcoxon signed ranked test is conducted to demonstrate significant differences amid QDPSKNN and state-of-the-art methods.

Real Time Mobile Ad Investigator: An Effective and Novel Approach for Mobile Click Fraud Detection

Today, mobile advertising is considered as the most effective medium to convey promotional messages to customers because of the excessive usage of mobile phones and tablets all around the world. However, this ecosystem has severely been affected by fraudulent activities due to a large sum of money circulated in the advertising industry. The term ad fraud is referred to as any kind of fraudulent activities that are executed by fraudulent users either a human or an automated script. The combat between researchers and fraudulent users never ends because more smarter strategies are being used by the fraudsters to bypass the significant number of detection and prevention solutions. The Real Time Mobile Ad Investigator-RTMAI is proposed as a software solution to address this problem where a novel supervised learning algorithm based on the hidden Markov model along with a rule engine have been proposed to classify fraudulent impressions in real time. Furthermore, RTMAI proposed a solution to address the class imbalance problem which is generic to most of the classification datasets. The experimental results show the significance of the proposed approach to classify the fraud or non-fraud clicks/events, impressions and even user sessions more confidently in real time.

A hybrid and effective learning approach for Click Fraud detection

Click Fraud is a fraudulent act of clicking on pay-per-click advertisements to increase the site’s revenue or to drain revenue from the advertiser. This illegal act has been putting commercial industries in a dilemma for quite some time. These industries think twice before advertising their products on websites and mobile-apps, as many parties try to exploit them. To safely promote their products, there must be an efficient system to detect click fraud. To address this problem, we propose a model called CFXGB (Cascaded Forest and XGBoost). The proposed model, classified under supervised machine learning, is a combination of two learning models used for feature transformation and classification. We showcase its superior performance compared to other related models, and make a comparison with multiple click fraud datasets with varying sizes.

The Effect of Feature Reduction in Click Fraud Detection: Review

It is almost impossible for online activities being without fraud. Online ads face a major threat represents by fake clicks which happen because of bots or some mischievous people. Several studies have solved the problem using machine learning algorithms. Some of them have solved only the problem of automatic click fraud (which carried out using bot), to classify physical or bot click. While many recent researches have detected click fraud problem in spite of clicks type. This paper presents a survey of methods used to detect fraud clicks on ads. It presents advantages, as well as disadvantages of each method, in general, Most recent studies in this field, have focused on features preprocessing before classification, because of the problems’ type which imposed existence many related features and this may lead to overfitting. So the solution is applying dimensional reduction algorithms, to get better results and avoid overfitting. Keywords: Click Fraud, dimensional reduction, features, Online advertising, pay_per_click. DOI: 10.7176/NCS/11-01 Publication date: July 31st 2020

AdSherlock: Efficient and Deployable Click Fraud Detection for Mobile Applications

Mobile advertising plays a vital role in the mobile app ecosystem. A major threat to the sustainability of this ecosystem is click fraud, i.e., ad clicks performed by malicious code or automatic bot problems. Existing click fraud detection approaches focus on analyzing the ad requests at the server side. However, such approaches may suffer from high false negatives since the detection can be easily circumvented, e.g., when the clicks are behind proxies or globally distributed. In this paper, we present AdSherlock, an efficient and deployable click fraud detection approach at the client side (inside the application) for mobile apps. AdSherlock splits the computation-intensive operations of click request identification into an offline procedure and an online procedure. In the offline procedure, AdSherlock generates both exact patterns and probabilistic patterns based on URL (Uniform Resource Locator) tokenization. These patterns are used in the online procedure for click request identification and further used for click fraud detection together with an ad request tree model. We implement a prototype of AdSherlock and evaluate its performance using real apps. The online detector is injected into the app executable archive through binary instrumentation. Results show that AdSherlock achieves higher click fraud detection accuracy compared with state of the art, with negligible runtime overhead.

Crowdsourcing for click fraud detection

Mobile ads are plagued with fraudulent clicks which is a major challenge for the advertising community. Although popular ad networks use many techniques to detect click fraud, they do not protect the client from possible collusion between publishers and ad networks. In addition, ad networks are not able to monitor the user’s activity for click fraud detection once they are redirected to the advertising site after clicking the ad. We propose a new crowdsource-based system called Click Fraud Crowdsourcing (CFC) that collaborates with both advertisers and ad networks in order to protect both parties from any possible click fraudulent acts. The system benefits from both a global view, where it gathers multiple ad requests corresponding to different ad network-publisher-advertiser combinations, and a local view, where it is able to track the users’ engagement in each advertising website. The results demonstrated that our approach offers a lower false positive rate (0.1) when detecting click fraud as opposed to proposed solutions in the literature, while maintaining a high true positive rate (0.9). Furthermore, we propose a new mobile ad charging model that benefits from our system to charge advertisers based on the duration spent in the advertiser’s website.

User click fraud detection method based on Top-Rank-k frequent pattern mining

A user click fraud detection method based on Top-Rank-k frequent pattern mining algorithm is presented to solve the click fraud problem appearing in current online advertising. Firstly, this method combines the click frequency of event samples, calculates the real evaluation score of click stream, and the click stream density function and evaluation score expression under multi-dimensional variables, and further obtains the time complexity of the next user’s click fraud process. Secondly, according to the Top-Rank-k frequent pattern, the process of click fraud detection algorithm is designed, and the click fraud user is analyzed and obtained. The results show that this method has good efficiency and correctness, and is superior to other similar algorithms.

Click Fraud Detection and Prevention System for Ad Networks

Click fraud detection consists of identifying the intention behind received clicks, given only technical data and context information. Reviewing concepts involved in click fraud practices and related work, a system that detects and prevents this type of fraud is proposed and implemented. The system is based and implemented on an ad network, one of the 3 main agents in the online ad environment, and for its validation, 3 servers were used, representing the publisher, the ad network with the system implemented and the announcer, and a bot that attempts to commit a click fraud.

An EM Algorithm for Click Fraud Detection

This paper is concerned with the problem of click fraud detection. We assume each visitor of a website carries a latent indicator, which labels him/her as a regular or malicious user. Information such as number of clicks, number of page views (PVs) and time difference between consecutive clicks are cooperated in our newly proposed statistical model. We allow those random variables to share the same distribution but with different parameters according to the visitor's type. An EM algorithm is then suggested to obtain the maximum likelihood estimator. As a result, click fraud detection can be implemented by estimating the posterior malicious probability of each visitor. Simulation studies are conducted to assess the finite sample performance. We also demonstrate the usefulness of the proposed method via an empirical analysis of a real life example on searchengine marketing.

Duplicate detection in pay-per-click streams using temporal stateful Bloom filters

Detecting duplicates in click data streams is an important task to fight against click fraud, which is the act of generating false clicks in internet advertising. Revenue generation advertising models, that charge advertisers for each click, leave space for individuals or rival companies to generate false clicks. The extent of click fraud's damage to online advertising has grown tremendously over the years. In this paper, we consider the problem of detecting duplicates in click data streams. Our solution uses a modified version of the counting Bloom filter. The temporal stateful Bloom filter (TSBF) extends the standard counting Bloom filter by replacing the bit-vector with an array of counters of states. These counters are dynamic and decay with time. We conducted a comprehensive set of experiments using synthetic and real world data. Results are compared with buffering techniques used in NetMosaics, a click fraud detection and prevention solution. Our results show that TSBF approach achieves 99% accuracy on duplicate detection, while keeping its space requirement a constant.