Insurance fraud continues to be a significant issue faced by insurers across the globe, resulting in huge financial losses per year and reducing the industry's efficiency. Insurance fraud, which involves overstated losses, faked claims, or any other form of deception, adds to operational expenses and increases the cost of policies for legitimate policyholders. Due to the difficulty of detecting insurance fraud, there has been an increased interest in analytics solutions that can deal with complex data, especially when it is scaled up (Maria Chousa Santos et al., 2024).
To study fraud based on a dataset, this study uses multiple approaches, which includes the identification of factors that are related to fraud, fraud prediction based on machine learning algorithms, and temporal trends of fraud. Each part of the methodology depends on each research question.
├── R_codes
│ ├── construction_project_raw.csv
│ ├── insurance_fraud_raw.csv
│ │ ├── Bhuvan_Final_Code.R
│ │ ├── DATA614_Final_Project_TS_bilal.Rmd
│ │ ├── insurance_fraud_analysis.R
│ │ ├── micro_spike_timeseries.R
│ │ ├── readme.md
│ │ ├── time_series_modelling.R
│ │ └── ...
│
├── brainstorming
├── seperate_report
The data set being utilized within this project is the Insurance Fraud Detection data set, and can be accessed via Kaggle via the link provided below: https://www.kaggle.com/datasets/arpan129/insurance-fraud-detection. This data set is a research carried out at Indian Institute of Management Calcutta, and comprises 1,000 records of insurance claims. There are 39 variables, 38 of which are independent while the last one (fraud_reported) is dependent. The dependent variable consists of binary values which take a value of either zero to indicate legitimacy, or one to indicate fraudulence. The data set comprises both numerical and categorical variables, including the nature of incidents (incident_severity), claim amounts (total_claim_amount), property damages, customer tenure amongst others. This is a classic example of fraud detection, where the minority class consists of fraudulent transactions, thus causing a problem of class imbalance which can impact models performance and evaluation.
- Question 1: What factors are most strongly associated with insurance fraud claims?
- Question 2: Can we accurately predict fraudulent insurance claims using machine learning models?
- Question 3: How do fraud patterns evolve over time, and what temporal trends can be identified in insurance claims?
- Youtube video: https://www.youtube.com/live/k3ftvLRAY5w
Temporal limitations in the use of the data set make it difficult to find patterns that are significant through time. While the limited data from early 2015 gives rise to interesting spikes in fraud cases in particular months, there are no other data that extend through time to allow one to examine the trends regarding insurance frauds and claims filings over a long period of time. Thus, to conduct an effective analysis concerning the temporal aspects, the data set must be expanded significantly.
Insurers who seek ways to improve their fraud detection systems will benefit from the results of this research in various ways. First, it is important to note that the study identifies several significant features of a case, including its level of severity, total claim value, and damage to property. These parameters could help insurers identify risky cases that require deeper examination. Secondly, it appears that the recall rate of the random forest algorithm indicates its potential use in the screening process to avoid missing any frauds while producing false alerts. Finally, claim attributes outweigh demographic characteristics, making it easier for firms to develop an unbiased system.
- du Preez, A., Bhattacharya, S., Beling, P., & Bowen, E. (2024). Fraud detection in healthcare claims using machine learning: A systematic review. Artificial Intelligence in Medicine, 103061. https://doi.org/10.1016/j.artmed.2024.103061
- Hamid, Z., Khalique, F., Mahmood, S., Daud, A., Bukhari, A., & Alshemaimri, B. (2024). Healthcare insurance fraud detection using data mining. BMC Medical Informatics and * Decision Making, 24(1). https://doi.org/10.1186/s12911-024-02512-4
- Maria Chousa Santos, Pereira, T., Mendes, I., & Amaral, A. (2024). Machine Learning for Insurance Fraud Detection. 56–65. https://doi.org/10.1007/978-3-031-51572-9_5
- Severino, M. K., & Peng, Y. (2021). Machine learning algorithms for fraud prediction in property insurance: Empirical evidence using real-world microdata. Machine Learning with Applications, 5, 100074. https://doi.org/10.1016/j.mlwa.2021.100074
- Arpan129. (2019). Insurance Fraud Detection [Data set]. Kaggle. https://www.kaggle.com/datasets/arpan129/insurance-fraud-detection
Group project is an important part of the class. Students will exercise formulating and solving problems using models and tools learned throughout the course. You will also practice group work and presentation skills, both of which are crucial for your careers in business. Group project grade consists of project write-up (intro, data, and recommendations), visualization/modelling, and presentation. There is also a grade for peer evaluations. The details of requirements for group project will be provided in D2L. Consult your ideas with the instructor early to help you better prepare your projects. You are required to present your projects (~15 minutes) to the class on the last day of class. The Group Project files should be submitted through D2L Dropbox
Students are required to submit white papers in groups on a data analytics topic. Most appropriate topics can be business applications of one or more data analytics models/tools of your choice. I suggest papers 2-3 pages long, including an intro on the models/tools and examples of business applications. Further details will be provided on D2L. Consult your ideas with the instructor early on to help you better prepare your papers. You are required to submit your papers on D2L and present your papers in short format (about 5 minutes) to the class on the same day as group projects (last day of class). The Group Paper files should be submitted through D2L Dropbox
