AbstractMigrant remittances have become significant in poverty alleviation and microeconomic development in low-income countries. However, the ease of conducting global migrant remittance transfers has also introduced the risk of misuse by terrorist organizations to quickly move and conceal operational funds, facilitating terrorism financing. This study aims to develop an unsupervised machine learning algorithm capable of detecting suspicious financial transactions associated with terrorist financing in migrant remittances. To achieve this goal, a structural equation model (SEM) and an outlier detection algorithm were developed to analyze and identify suspicious transactions among the financial activities of migrants residing in Belgium. The results show that the SEM model classifies a significantly high number of transactions as suspicious, making it prone to detecting false positives. Finally, the study developed an ensemble outlier detection algorithm that comprises an isolation forest (IF) and a local outlier factor (LOF) to detect suspicious transactions in the same dataset. The model performed exceptionally well, being able to detect over 90% of suspicious transactions.
2.7. Reachability Distance The reachability distance of an object, p, with respect to object O is defined as follows: reach distk(p, O) = max{k distance(O), d(p, O)} Figure 2 shows an illustration of the reachability distance with k = 4. When the object, p, is significantly far from O, the reachability distance is measured as the actual distance between the object p2 and O, which is depicted by the distance, reach distk(p2, O), in the figure below. However, if the object, p, is sufficiently close to O or located within the k-distance radius, then the k-distance of O becomes the reachability distance. 9 of 19 (9) J. Risk Financial Manag. 2024, 17, 181 Figure 2. An illustration of the reachability distance with k = 4 (adapted fromBreunig et al. (2000)).
id: e80f6c914080afc9e04a0002d88f0deb - page: 9
2.8. Local Reachability Density The concept of density is defined by two parameters: the minimum number of objects (MinPts) and the parameter specifying volume. They are crucial in determining the density threshold for a density-based clustering algorithm; that is, the objects that exceed a particular density threshold are connected. Therefore, to detect density-based outliers, it is necessary to compare the different cluster densities. The local reachability density of p measures the volume that determines the density in the neighborhood of an object, p, and is defined as follows: LrdMinPts(p) = (cid:20) ONMinPts(p) reach distMinPts(p,O) |N MinPts(p)| (cid:21)1 Thus, the local outlier factor of p is the average of the ratio between the local reachability density of p and its MinPts-nearest neighbors, and is defined as follows: LOFMinPts(p) = ON MinPts (cid:16) lrdMinPts(o) lrdMinPts(p)
id: 3d76e2e7fbead10396ee97be7cfcff11 - page: 10
N MinPts(p) (cid:17) If the local outlier factor for a data point exceeds 1, then it is considered an outlier (Breunig et al. 2000). 2.9. Proposed Outlier Detection Algorithm The anomaly detection algorithm used for the detection of suspicious migrant remittances is an ensemble model developed from the combination of the isolation forest and the local outlier factor (LOF-IF). As previously highlighted in Section 2.5, the isolation forest algorithm is sensitive to detecting global or extreme outliers but performs poorly in identifying local outliers in data clusters of varying densities. In contrast, the local outlier factor algorithm proves to be robust in detecting such local outliers at the expense of a high computational cost when compared to the isolation forest. The two algorithms will essentially cancel out the drawbacks of each other, simultaneously improving the model performance and lowering the time complexity (Cheng et al. 2019).
id: d719f647f03ab37e2b0edbe52ed46c15 - page: 10
This ensemble algorithm was previously utilized in the existing literature. Wang and Xu (2019) combined the isolation forest and the local outlier factor to improve the detection of anomalies found in concrete mixtures. The paper proposed an isolation forest algorithm based on a sliding window technique for the local outlier factor. The sliding window creates a window-size data storage that stores the data points computed from the isolation forest. The local outlier factor algorithm then uses a threshold to calculate the outlier score 10 of 19 (10) (11) J. Risk Financial Manag. 2024, 17, 181 from the input data obtained from the sliding window. Data points exceeding the threshold value would be considered outliers.
id: f5158e4d6727370619245b8644f8c751 - page: 10