LINK ANALYSIS ON SOCIAL ENGINEERING

Abstract

Modern-day crime often involves multiple actors working in coordination across regions or even continents. Criminals might use encrypted communication, proxies, and false identities to avoid detection. This complexity makes it difficult for investigators to isolate suspects or determine their roles without sophisticated analytical techniques. Link analysis helps overcome these challenges by enabling investigators to model and understand the structure and flow of criminal activity. For example, when investigating a criminal gang involved in narcotics, investigators can use link analysis to map out connections between known suspects, trace their call histories, track their financial transactions, and link them to locations or events (eg. Drug seizers). This thesis analyses the challenges faced by link analysis methods and how different algorithms can improve it. Algorithms like Spizella, SDHO based detection are formulated in this research to overcome the existing challenges in link analysis. The first study in this thesis analyses and finds gaps in different Link Analysis tools and techniques. Link analysis is a technique of data mining that is especially used to detect useful and interesting patterns. The first challenge in link analysis is to reduce the graphs into manageable portions. Identifying the interesting relationships and deciding to reduce the graphs is important challenge. Therefore, determining how to apply the link analysis techniques to detect abnormal and suspicious behaviour is needed. The second study proposes a model that, through rule-based deduction, data transformation, and FP-Growth algorithms, detects patterns of influence between states. This approach clusters data by crime types, finding that criminal activities in Alaska impact Washington, while Alaskan policies have a more influence Washington. These findings align with broader national trends: recent FBI data indicates significant decreases in violent crime nationwide in 2023, underscoring the impact of policy adjustments across states. The third study proposes a Spizella swarm based BiLSTM classifier is used for the detection of crime rate in this research. Faster convergence is a crucial factor and this faster convergence is achieved by the proposed Spizella swarm optimization. BiLSTM classifier effectively identified the crime rate and the BiLSTM performance is boosted by the Spizella swarm optimization where the escaping characteristics of Spizella improve the convergence iv and help in attaining desired results. Measuring the metrics values for accuracy, sensitivity, and specificity demonstrates the effectiveness of the proposed method. The fourth study introduces a Sheep Dog Hunt Optimization enabled Knowledge- Enhanced Optimal Graph Neural Network classifier (SDHO-KGNN) approach for detecting fraudulent calls accurately. The effectiveness of the proposed SDHO-KGNN approach is achieved through the combination of the power of graph representation learning with expert insights, which allows the proposed SDHO-KGNN approach to capture complex relationships and patterns within telecom data. Additionally, the integration of the SDHO algorithm enhances the model performance by optimizing the discrimination between legitimate and fraudulent calls.