A user-centric recommendation system for anomaly exploration in the healthcare domain

In the era of data-driven decision-making, detecting and analysing anomalies have become increasingly critical, particularly within the healthcare sector. In this context, enhancing efficiency, accuracy, and the early identification of irregularities is essential. Given the complexity, and growing volume of healthcare data, there is a pressing need for intelligent systems that not only identify anomalous patterns, but also support users in effectively navigating vast amounts of information. This thesis investigates the integration of recommendation system frameworks within an outlier analysis platform on healthcare data, intending to improve user experience and streamline investigative processes. Three distinct recommendation approaches were developed and evaluated: (i) Rank Aggregation, designed to improve the prioritization of elements on the platform’s homepage by refining ranking mechanisms; (ii) Similar Product Recommendation, which assists users in discovering related outliers through a clustering-based approach and a nearest-neighbor ranking algorithm; and (iii) User-Based Personalized Recommendation, leveraging user interaction history to generate tailored suggestions via an autoencoder-based embedding model and a binary classifier. This work holds significant implications for the healthcare domain, where identifying anomalies in medical and administrative data can lead to better resource allocation, fraud detection, and improved patient outcomes. By integrating recommendation techniques, the system empowers users with intuitive tools to explore and understand critical insights. The emphasis is placed on a user-centric approach, ensuring that recommendations align with individual search behaviors, preferences, and investigative needs. This research lays the foundation for intelligent, user-centered outlier detection in healthcare data, offering a valuable tool for professionals navigating complex datasets and improving decision-making efficiency.