Predictive Risk Modeling in P&C Insurance Using Guidewire DataHub and Power BI Embedded Analytics

2025-07-07T14:00:49+08:00

P&C insurers are increasingly pressured to identify and effectively predict risk. While traditional methods, such as actuarial models and manual assessments, are effective for identifying patterns in large-scale policy and claims data, they struggle to capture complex patterns, like resistance curves. This paper examines how predictive risk modelling can be implemented in practice using Guidewire DataHub and Power BI Embedded Analytics. Power BI is used for interactive visualization and real-time decision support, whereas Guidewire Data Hub is utilized as a centralized platform for storing and managing structured insurance data. It utilized structured data from claim history, underwriting attributes, policy details, and customer profiles to build a predictive model. Machine learning algorithms, such as Random Forest and Logistic Regression, were then applied to classify policyholders as High, Medium, or Low risk after preprocessing and feature selection. Standard metrics (accuracy, precision, recall, ROC-AUC) were used to evaluate model performance. The Random Forest classifier achieves an accuracy of 84% and identifies high-risk profiles most effectively. It then integrated these predictions with Power BI dashboards, allowing underwriters and analysts to explore risk at both the individual and portfolio levels. The study illustrates how building data platforms that integrate machine learning and embedded analytics facilitates more innovative underwriting, fraud detection and pricing. In a competitive, data-driven insurance environment, the ability to turn raw insurance data into actionable insights provides significant operational and strategic value.

Security and Privacy Testing Automation for LLM-Enhanced Applications in Mobile Devices

2025-07-23T13:46:51+08:00

The integration of large language models (LLMs) into mobile applications introduces new vectors for security and privacy vulnerabilities. This study proposes an automated framework for systematically testing LLM-enabled mobile apps, focusing on identifying potential threats such as prompt injection, data leakage, unauthorized access, and adversarial manipulation. The approach combines dynamic analysis, static code inspection, and machine learning-based anomaly detection to evaluate app behaviors in real-time. Our method ensures scalability and efficiency across diverse mobile platforms and LLM configurations. Results demonstrate significant improvements in detection rates and response times compared to conventional manual testing. This work aims to bridge the gap between AI innovation and secure mobile deployment, promoting trust in AI-integrated ecosystems.

Международный журнал по сетям и безопасности

Predictive Risk Modeling in P&C Insurance Using Guidewire DataHub and Power BI Embedded Analytics

Security and Privacy Testing Automation for LLM-Enhanced Applications in Mobile Devices