Counterfeit product grouping - a cluster ensemble approach

Abstract: he need for the development and use of efficient and reliable Anticounterfeit techniques in the manufacturing industry and online trading has recently increased tremendously. Flooding counterfeit products in the international market have several adverse social and economic impacts globally. Manufacturing Companies are now forced to invest a lot of money in developing Anti-counterfeit techniques. Most of the methods conventionally employed to prevent counterfeiters from producing and distributing fake products are either obsolete or inefficient. The counterfeiters adapt themselves to the new situation fast. They use newer technology to create products, which is hard to impossible to identify by human inspection. The attempts to develop modern techniques to overcome the threat raised by counterfeiters have been identified as an important research activity. These techniques should be capable of identifying a fake product in a fast and efficient manner with a relatively smaller investment. An effective combination of cloud computing facilities and machine learning algorithms can be advantageous in this area. Therefore, this thesis aims to design and launch a reliable system to pinpoint counterfeit products using machine learning methods. This thesis uses a cloud platform for image data collection, which makes the system more flexible. Usage of the cloud also brings down the overhead of the processing system. The model developed in this thesis can be merged into the back-end of the software to identify the product’s genuineness even using a mobile phone and turns out to be a boon from the end user’s point of view. The starting point of any process related to counterfeit product identification is based on the analysis of data pertaining to the previously identified counterfeited products (historical data). These product data are not generally labeled. Regarding the data we usually have previous information on the categories like clothes, watches, vehicles, etc. Hence unsupervised machine learning techniques are being used. The machine identifies in which group the incoming data falls. If the data forms part of any of the previously determined clusters it is recorded as a counterfeit product. For example, if the data falls in the watch cluster it is identified as a forfeited watch product. The initial hypothesis is that an unsupervised machine learning technique (cluster algorithm - ensemble cluster) could be used in the Anti-counterfeit product detection process through analysis of historical data. It has been proven to be reasonably accurate based on our experimental investigations on sample data. In this study, the cluster ensemble method is used for the first time. Merged with a front-end obile application the system can be converted into a reliable userfriendly product. The clustering methods available in Python with the scikit-learn machine learning library are used in this project.