http://hdl.handle.net/20.500.14044/33843 2026-07-26T13:04:18Z http://hdl.handle.net/20.500.14044/31961 Segmentation of Electricity Consumers Using Clustering Sarnovsky, Martin; Bednar, Peter The work presented in this paper is focused on customer segmentation based on the electricity demand by clustering methods. The main goal was to cluster the customers of a local electrical energy distribution company into groups with similar characteristics of electricity consumption based on annual data from smart metering systems. Such customer groups can be used to ease the understanding of the differences in behaviour of individual customers, and further can be used in targeted marketing or other machine learning tasks, such as consumption prediction for specific groups. The work followed the CRISP-DM process model, a commonly used methodology for the application of data analytics in businesses. In the paper, we briefly describe each phase of the methodology and present the most important outputs. The resulting customer segments are described and interpreted, and visualizations of clusters were provided, which help to better understand the behavior of customers. 2025-01-01T00:00:00Z http://hdl.handle.net/20.500.14044/31960 Factors Influencing the Intention to Use Robo- Advisors: A Hungarian Perspective Molnár, László; Nagy, Szabolcs; Hajdú, Noémi This study aims to investigate the factors influencing the intention to use robo- advisors, based on an extended Unified Theory of Acceptance and Use of Technology (UTAUT) model, incorporating trust and perceived risk as new elements, alongside Artificial Intelligence attributes. To test our conceptual model, we conducted a survey in Hungary in 2024, with 249 respondents completing our online questionnaire. The model and hypotheses were evaluated using structural equation modeling (SEM). The results indicate that the intention to use robo-advisors is most significantly influenced by performance expectancy, trust, social influence, and facilitating conditions. Among the AI attributes, perceived intelligence stands out, exerting an indirect effect on the intention to use through the aforementioned factors. A limitation of our study is its’ geographical focus on Hungary, restricting the generalizability of the findings to potential Hungarian users. Additionally, we were unable to investigate actual usage due to the currently low service penetration. Understanding the factors that influence the preference for automated investment management solutions over traditional advisors is essential for marketing managers in fintech companies to devise effective client acquisition and retention strategies. The findings highlight the importance of trust, security, and digital literacy. Addressing these factors is vital for maximizing the benefits and mitigating the risks associated with AI in financial services. The originality lies in its integrated examination of perceived intelligence and anthropomorphism within an extended UTAUT model, uncovering their combined effects on the intention to use robo-advisors. 2025-01-01T00:00:00Z http://hdl.handle.net/20.500.14044/31958 NTC-CIL: Characterizing and Classifying Encrypted Network Traffic using Class- Incremental Learning Gudla, Raju; Vollala, Satyanarayana; Amin, Ruhul; Abdussami, Mohammad In the field of network security and management, accurately identifying and managing encrypted traffic is essential for mitigating potential attacks and optimizing resource usage. However, conventional methods often underperform in adapting to new traffic classes, require more manual intervention, time-consuming, and resource-intensive. These limitations reduce system performance and increase vulnerability issues. Conventional models also face scalability issues and are prone to catastrophic forgetting, where previously learned traffic patterns are lost as new ones are introduced, leading to reduced classification accuracy over time. To address these challenges, we propose a novel method: Network Traffic Classification using Class-Incremental Learning (NTC-CIL). NTC-CIL combines a random forest classifier with the Learning without Forgetting (LwF) method, an incremental learning method based on knowledge distillation. This approach enables the model to retain previously learned patterns while incorporating new traffic classes, including encrypted and evolving types. As a result, NTC-CIL can continuously adapt to unfamiliar network traffic without retraining from scratch. Experimental evaluations demonstrate that NTC-CIL outperforms existing techniques by achieving an accuracy of 97%. This marks a significant advancement for network security, offering a scalable and adaptive solution capable of detecting new threats in dynamic traffic environments. 2025-01-01T00:00:00Z http://hdl.handle.net/20.500.14044/31956 Comparative Analysis of AI models ‒ Using AI-supported Qualitative Data Analysis for Interview Analysis Sterczl, Gábor; Csiszárik-Kocsir, Ágnes This study aims to comparatively evaluate the performance of currently popular Artificial Intelligence (AI) models in supporting qualitative data analysis, specifically focusing on the coding and hypothesis validation of interview transcripts. We investigate how models from OpenAI, Google Gemini, and Anthropic perform in these tasks compared to traditional manual analysis and established CAQDAS tools. Utilizing transcripts from three exploratory interviews, the methodology involved applying each AI model and selected CAQDAS tools to generate codes and quantify references based on predefined research objectives and a set of established codes. Key findings reveal significant variability in the ability of different AI models to accurately identify and quantify relevant data segments, with some models demonstrating greater efficiency and the capacity to suggest novel, relevant categories not initially identified through manual analysis (e.g., external influences, roles, and responsibilities). Conversely, instances of inaccuracies, such as hallucinated quotes, were observed in other models. The study highlights that while AI offers substantial potential for increasing the efficiency and objectivity of qualitative analysis, its effectiveness is highly dependent on the specific model used and necessitates critical human oversight and validation. The implications underscore the importance of a hybrid human-AI approach in qualitative research, emphasizing careful model selection, robust data management protocols, and continuous attention to ethical considerations, particularly regarding data privacy and algorithmic bias. 2025-01-01T00:00:00Z