DEVELOPMENT OF ENHANCED BAYESIAN MODEL FOR DETECTION OF COVERT MEMBERS IN CRIMINAL NETWORKS USING TELECOMMUNICATION METADATA

Abstract

Crime has become a global challenge in recent times. The phenomenon has become a difficult task that military war-fare approach alone can address effectively without intelligence. Criminal intelligence involves gathering data on criminal activities and participants for preparing deplorable strategies and interventions. Social Network Analysis (SNA) offers supportive tools for analysing Organised Criminal Groups (OCGs) and identifying important nodes with conspicuous relationship as its priority. SNA-based techniques arrived at key players in criminal network with nodes that have high SNA metric values. Apart from datasets challenge, SNA is a weak scheme for key players in OCGs because conspicuous links raise susceptibility of vibrant participants while silent key actors are concealed. Also, status of key actors in OCGs are unrelated with SNA metrics. Scatter-graph of vulnerability and strategic positions was devised to mitigate unrelatedness of SNA metrics for detection of key players in Criminal Social Network (CSN). The scheme identifies actors that have both high vulnerability and high strategic position values at the same time. This is synonymous to Influence Maximization (IM) – set of nodes that have high influence. Silent key players or legitimate actors in adversary network still remain unresolved. Missing node concept works towards set of nodes not known initially as part of a social criminal group. It has high affinity for well-connected nodes than marginal nodes. Node discovery scheme unravels latent structure behind key players within CSN. The scheme pinched on multiple sources of data about a criminal group yet legitimate actor are not captured. Inference approach offers probability-based prediction for detecting covert nodes yet only well-connected nodes with conspicuous relationships are still identifiable. The development of Enhanced Bayesian Model aimed at predicting key players like financial aiders and ammunition suppliers with evasive attitudes. It was conceived towards inherent problem of erratic behaviour and structural equivalence abating key-players from theoretical graph-based. Bayesian model and Recursive Bayesian Filter (RBF) algorithm were combined to have Enhanced Bayesian Network Model (EnBNM) with RBF to lower error rate and improve prediction. EnBNM scheme re-ranks participant’s attribute by assigning inference to nodes base on conditional probability of Bayesian model. EnBNM’s algorithm was validated using ground truth and SNA-Q model adopted for classifying Criminal Profile Status (CPS). EnBNM was tested using dataset of participants in November 17 Greece revolutionary group - (N’17) and data of participants in September 11 Al-Qaeda terrorist group - (9/11). For N’17 dataset, EBNM detected all alleged and convicted leaders. Additional two actors were detected who had the same CPS with convicted leaders. EnBNM also detected marginal actors; participants with high tendency to evasion. Out of four (4) detected fugitives, two of them belong to the first-generation leadership (G) faction. For 9/11: nine (9) out of nineteen (19) central participants detected by EBNM have the same CPS with convicted leaders. It means that seven (7) more actors are detected as additional key players by EnBNM that previous models did not detect. Six of these actors detected are conspirators. A financial aider to the group was detected among fugitives. The results corroborate that terrorist organisations are self-organised with decentralised key players as a measure to minimize effect of security perturbation. The simulation results showed that the court judgement of the N’17 group was 40% in error as additional two actors were detected by EBNM apart from the three convicted leaders by court. It shows that support of intelligence is highly needed for effective disruption of OCG and terrorism. The EnBNM algorithm also detected over 80% of legitimate actors - less vulnerable participants in the 9/11 terrorist group and has 59.09% accuracy score in detection of conspirators.