الفهرس 
Chapter One: IntroductionOnly 14 pages are availabe for public view
 |  | from | 102 |  |  |
| | | from | 102 | | |
Abstract
”Recent years have witnessed the emergence of novel ideas and concepts of big data to cope with the remarkable rise and accumulated amounts of data in many business sectors. Meanwhile, the remarkable growth of internet uses and social networks has even added insufficient data and enforced more challenges facing the conventional data processing systems. The insurance sector, the target case in this research, is characterized by its information-driven feature as it generates and accumulates huge amounts of records, both structured and unstructured which makes traditional data processing techniques unable to handle. On the other hand, Social insurance is identified as the main focus field for our research. Social insurance, in brief, refers to policies, procedures, data processing, and decisions related to an individual’s protection against risks such as retirement, death, or disability. This dissertation presents and surveys some recent research related to big data mining and analytics, focusing on insurance claims, finance in insurance industries, and pensions. Moreover, the dissertation illustrates a brief review of social insurance in Egypt while also justifying why social insurance data is considered big data. The dissertation proposes a framework that comprises social insurance big data from the data extraction stage to the use cases stage in social insurance with guidelines about how to apply it in Egypt. It’s claimed that the proposed framework could be an effective start-up point to enhance social insurance services, help the insurers and actuaries to make good and fast decisions and it also enables more insights into the social insurance data. The dissertation contributed a unique dataset collected from real Egyptian social insurance data with implementation details of the different phases are given in the framework presented which starts with the pre-processing steps on the dataset by using some famous methods such as replacing missing values with mean, standardization, and outlier/extreme values. Then applied unsupervised algorithm k-means and implemented four algorithms of supervised learning algorithms: Naive Bayes, Decision Tree, Support Vector Machine (SVM), and CN2 Rule Induction. The dataset split by 80% for training and 20% for the test to measure the performance of the four applied algorithms. The experiment illustrated many useful results, in clustering, the results showed a silhouette score of 0.790 with two clusters in the dataset features. Also, in classification, this research presents the comparison results between the aforementioned four supervised learning algorithms. The classification results showed that the CN2 inducer is the most successful algorithm in the classification experiment of the social insurance dataset with a high accuracy percentage up to 95.04% with an F1 score of 0.9445, precision 0.9464, and recall 0.9504. Consequently, this research verified the added value of the proposed model to predict whether any individual in the social insurance scheme is insured or uninsured.