Browsing by Author "Wambu, Anthony Mwiti"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    An Ensemble Feature Selection Model with Machine Learning Model for Detection of Fraudulent Motor Vehicle Insurance Claims
    (Kenyatta University, 2025-05) Wambu, Anthony Mwiti
    Insurance companies are continuously inventing new competitive insurance products in order to enlarge their market share. This has continuously created opportunities for insurance fraud as well. Despite the insurance industry having extensive motor vehicle policy data and claims information, fraudulent claims remain a significant challenge in motor vehicle insurance. Proper analysis of this data can result in development of more efficient methods for identifying fraudulent claims. The challenge lies on how to extract valuable insights and knowledge from this data. This is because insurance datasets inherently include noisy features or low-quality subsets of data. This study used feature selection techniques to select relevant features from motor vehicle insurance claim dataset. The selected features were then used in training machine learning model. The machine learning model consisted of multiple machine learning algorithms whose individual prediction results were combined by use of a voting method. This helped to improve classification performance. Machine learning model’s performance with feature selected dataset and with full dataset was then evaluated using recall, precision and F1- score. The results indicated that the model trained with feature selected dataset performed better than the model trained with full dataset attaining higher values in recall, precision and F1-score. This indicated improved capability in minimizing false negative and improved overall effectiveness in fraud detection. For feature work the model developed for detecting fraudulent motor vehicle insurance claims can be enhanced by integrating machine learning techniques with nature-inspired optimization algorithms. This will help in better handling of extensive datasets and result to development of more rapid and effective models for identifying false claims.
  • Thumbnail Image
    Item
    An Ensemble Feature Selection Model with Machine Learning Model for Detection of Fraudulent Motor Vehicle Insurance Claims.
    (TIJER-International Research Journals(TIJER), 2024-04) Wambu, Anthony Mwiti; Araka, Erik
    Insurance companies are continuously inventing new competitive insurance products in order to enlarge their market share. This has continuously created opportunities for insurance fraud as well. In motor vehicle insurance, fraudulent claims continue to be a big challenge despite the insurance industry having vast amounts of motor vehicle insurance policy data and claim’s data. Proper analysis of this data can help to develop a more efficient way of detecting fraudulent claims. The challenge is how to extract insightful information and knowledge from this data since by their nature insurance datasets contain noisy features or subsets of data which are of poor quality. In order to achieve an effective machine learning model, one needs to choose the right set of features of data in the pre-processing step. Including noisy and less important features has proven to affect the performance of most of the existing machine learning models being used in the insurance companies. With aid of proper and effective feature selection techniques machine learning models that uses only relevant features of data can be developed in order to detect fraudulent insurance claims effectively. These models can be employed in insurance industry to aid in detecting fraudulent motor vehicle insurance claims. This will result in reduction of loss adjustment expenses and also improve customer satisfaction. Although there exist several other methods and ways of data preprocessing this study employed an ensemble multiple filter feature selection method. This study involved development of motor vehicle fraud detection model using multiple algorithms whose results were combined by use of a voting method.