My thesis, titled "User Action Prediction System for Automated Home Based on Association Rules and Ontology", introduces a predictive model designed for smart home environments. This model addresses the limitations of current predictive systems, which often focus exclusively on frequent user actions while neglecting infrequent but potentially significant behaviors. By enhancing infrequent patterns and integrating them into the system, the research aims to improve prediction accuracy and make home automation systems more responsive to diverse user needs.

The model employs a hybrid approach, combining the FP-Growth algorithm and ontology-based methodologies. The FP-Growth algorithm is utilized to mine both frequent and infrequent patterns from user action data, while the ontology framework provides contextual relationships using time and location information. A custom formula integrates the contributions of time and location percentages, determining whether low-confidence rules are promoted for prediction. This structured approach allows the system to predict future user actions, making the smart home more adaptive and intuitive.

Developed using Java and supported by RapidMiner Studio, the system analyzes data collected from user routines over a three-month period. The process involves identifying patterns in daily activities and enhancing rules with confidence values below 80%. By applying the ontology promotion mechanism, the model successfully predicts actions that are otherwise overlooked in traditional systems. A weight distribution of 70% for time ontology and 30% for location ontology yielded the most realistic and reliable results, achieving an overall accuracy of 76.35%.

However, the model has some limitations. It was tested on a single lifestyle (employee lifestyle) and focused primarily on specific devices like lights within a fixed home size. These constraints limit its applicability across broader contexts. Despite these limitations, the research highlights promising results and offers a foundation for further improvements in predictive home automation systems.

Future directions for this work include extending the model to incorporate more devices, such as TVs and heating systems, and applying the methodology to other environments like offices and factories. Additionally, integrating user feedback could further refine the predictions, while including weekends in the analysis would broaden the model’s applicability. This thesis presents a novel approach to user action prediction in smart homes, effectively leveraging the strengths of association rules and ontology to enhance automation and user experience.