Fuzzy Approach to Identity Resolution

This is a pre-recorded presentation that is a part of 4th Student and Staff Research Conference: 'Small Steps Matters - Research towards a better world' organised by London Metropolitan University between 4th and 5th July 2023. Fuzzy Approach to Identity Resolution by Asif Nawaz. Identity resolution is to match true identity from a pool of duplicates or similar identities and it is a crucial and difficult task for law enforcement agencies globally. In big data, it is impossible to match true identity manually by a human expert but machine learning techniques can be helpful to tackle the problem. Using a supervised learning approach can achieve this but relies on training data and human experts. An unsupervised learning approach overcomes these issues as it is dynamic and no human expertise is required to execute tasks and provide training samples to the system to get the desired results. This paper proposes a fuzzy approach to identity resolution using unsupervised learning with fuzzy string similarity metrics for matching true identity. In the proposed model, the string similarity techniques are used where the Soundex technique has been modified to produce better results alongside the Jaro-Winkler technique. An iterative search uses a combination of Soundex and Jaro distance techniques to calculate the aggregated score for each name and search for target identity using a combination of different attributes. To group the searched records, a clustering technique is used with a dynamic generation of the number of clusters instead of using a fixed number. The Mean-Shift algorithm is utilised to provide the dynamic number of clusters based on the final dataset records retrieved during the iterative search. The overall matching performance using the aggregated score and matching categories is increased by only processing records that have similarities rather than processing irrelevant records. This approach results in a better and more effective identity resolution than other methodologies. The research can further progress by enhancing the record linkage using weighted match records with the neural network techniques. #londonmetresearch #LMUSmallStepsMatter #identityresolution #cybersecurity #machinelearning londonmet.ac.uk/research https://SmallStepsMatter-LMUStudentSt...