Designing Of Fraudster App Detector

Various approaches have been proposed to design fraudster mobile apps detector in Google play store. In this approach, malware was detected with the anonymized file submissions by using polonium algorithms. They used graph mining and inference techniques to detect the malware by constructing huge bipartite graph. The files with low repuatation are considered as malware. Online reviews is important for the users to buy the products on Internet, E-commerce websites, apps, etc. But many fake reviews are provided to mislead the users to buy a non reputated product. This method uses Network Footprint Score, a measure which identifies a product which is targeted by the spammers and then introduces a roup strainer approach to cluster the spammers. They use an unsupervised approach to find the spammer groups.

In another approach, User-generated online reviews are used by the target users to buy a product. But at times, fake reviews are given by fraudulent reviewers. They have proposed an approach which automatically detects the fake reviewers by using classification algorithms. This approach computes scores for reviews (whether fake or true), users (whether honest/fraud) and products (whether good/bad). Based on these scores the fake reviewers are detected.

In another paper, a new method called PUMA was used to detect the fake and malicious apps. It used machine learning techniques and extracted the permissions given for the applications. This method is used for extensive analysis. This method can be used for extensive dynamic analysis to detect the malware affected apps. The fraudulent app developers try to increase the ranking while searching the apps by providing fake reviews, etc. This happens because the popularity of the app and also provides financial benefits to the app developer. The main limitation of the existing approach is that the evidences of fraudulent behaviour are difficult to be obtained in a particular time due to which a reputated app will get affected. It’s not suitable for extracting fraud evidences at a particular given time period. Hence existing system does not accurately detect fraud effectively and not able to come up with a solution to solve the user’s problem The three evidences such as higher rating, ranking and good reviews are aggregated to finalize the best results will give the user to identify the fraud and fraudfree apps in the playstore. By analysing the fraud apps, user can safely download the best app that was recommended according to the specific category.

So this problem can be overcome by the solution which was suggested for the fraud detection in the apps in the playstore. An app which has higher rating, ranking and good reviews may attract more users to download and can also be ranked higher in the leader board. The rating, ranking and reviews are not always real to believe. Some fraudulent developers boost their apps dishonestly. The Google playstore fraud app is detected by aggregating the three evidences such as ranking based, co review based and rating based evidence. Thus by aggregating entire activities of leading apps, it can achieve accuracy in classifying standard datasets of fraudulent and legitimate apps. This paper proposes a framework for detecting fraudulent apps in Google Play Store. An incremental learning approach is proposed. The apps evidence such as rating, ranking and review evidences will be integrated by an unsupervised evidence-aggregation method for evaluating the mobile Apps. Here we have implemented incremental learning approach to effectively characterize the large dataset and to provide better aggregation. The aim of incremental learning is for the learning model to adapt to new data without forgetting its existing knowledge; it does not retrain the model. Some incremental learners have built-in some parameter or assumption that controls the relevancy of old data, while others, called stable incremental machine learning algorithms.

The Porter stemming algorithm is a process for removing the commoner morphological and inflexional endings from words in English. The five steps include normalisation, case folding, lemmatization, morphology and opinion words. The Google playstore fraud app is detected by aggregating the three evidences such as ranking based, review based and rating based evidence. The dataset from Google AppStore is obtained by crawling on the appstore. After obtaining the dataset, the first step is data preprocessing that involves transforming raw data into an understandable format. Preprocessing involves transforming the large set of data into table format which can be able to list the details of the particular app. These data are used for the decision making.

These information are used for the subsequent steps of analysing the fraud apps. The leading sessions of the app is found based on the popularity of the app, i.e. , based on a period of time. The previous data and current data are considered for mining leading sessions. This helps to extract fraud evidences. The historical data may consist of different ranking phases like rising, maintaining and recession phase. Even frequently downloaded apps may be fraudulent app, hence three evidences are considered to find the fraudulent app. The ranking fraud is located by mining the active periods of the corresponding app. The differences in the review dates are also considered. Rating for the published app can also be used which is determined by the ratings given by the downloaded user. If there is any anomaly patterns ratings found during active periods when compared with the historical ratings, it is also considered for constructing rating based evidences.

The mean value is taken for the segregated datasets. These helps to detect the fraudulent apps. Fake reviews are also detected by detecting local anomaly of reviews in leading sessions. Stop word removal and stemming is performed on the reviews to obtain the keywords of the reviews given by the user, after which the term frequency for each obtained words are calculated.

The similarity scores for the frequently occurs words are calculated and these scores are used to find the fraudulent apps by considering rating as well as ranking scores, ie. , these three evidences or scores are combined to find the fraudulent apps. Then the user is suggested to choose the best apps by avoiding fraudulent apps.