A Constrained and Cardinality Preserved Optimal Graph Attention Network for Fraud Detections in Insurance Claim
Bibliographic record
Abstract
Insurance companies face the pervasive issue of insurance fraud on a daily basis, especially in health and auto insurance. Such fraud can have a long-term effect on insurance premiums and significant losses for insurance firms. Insurance fraud research is currently focused on fraud detection. Traditional detection models consider only claims data, whereas it is crucial to consider the relationship between various parties involved in the claims data. To tackle this, Graph Attention Network (GAT) has emerged, which embeds the parties' relationships as a graph and learns the interdependencies of claims' parties to identify fraud. However, it suffers from overfitting and over-smoothing problems due to the imbalanced data problem. Poor discriminative power in GAT is caused by attention-based aggregation not taking cardinality information into effect. In addition, preconditioners affected the Adam optimizer, resulting in lower test accuracy. Thus, this article aims to address these challenges and enhance the efficiency of detecting insurance fraud claims in the vehicle and medical industries. First, a new Wholesome Oversampling Conditional Generative Adversarial Network (WOCGAN) model is proposed to address data imbalance problems. It integrates the Ordering Points to Identify the Clustering Structure (OPTICS) scheme and CGAN to remove borderline minority samples and oversample wholesome samples. An attribute extraction process is then performed to extract relationships between attributes and build a graph. Then, a new Constrained and Cardinality Preserved Optimal GAT (CCPOGAT) model is developed for fraud claims detection. In this model, a Cardinality Preserved Attention Strategy (CPAS) is proposed to enhance the attention strategy in GAT, capturing cardinality information during aggregation and distinguishing different multisets in the graph. A margin-based constraint on CPAS is introduced based on the class boundary to prevent overfitting and over-smoothing of GATs. Furthermore, a Hybrid Adam and Natural Gradient Descent (HAdaNGD) algorithm is adopted to optimize the architecture and improve the test accuracy of GAT. Finally, the proposed WOCGAN and CCPOGAT models are evaluated using different datasets to validate their efficiency compared to earlier models.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".