Publicação
Uncertainty and incompleteness handling in context-aware systems
| Resumo: | Ambient Intelligence (AmI) solutions are capable of acting autonomously to benefit human beings. This field of investigation is directly related to other domains, aiming to improve user experience by developing context-aware applications. Context information is naturally dynamic, uncertain, or incomplete. Thus, it is crucial to study ways to handle problems in context data to build applications that run autonomously. The main objective of this Ph.D. thesis is to identify elements that result in the characterization of uncertain contexts. For that, sources and types of uncertainty in intelligent environments were identified. A framework with three layers to validate context data was proposed. The first one defines an attribute grammar to ensure the structure of the data. Restrictions (expressed as logical formulas) associated with the rules of the grammar enforce the system to work with well-defined values. Pieces of context data should have quality, be relevant and be complete to be used by a context-aware system. Thus, the second layer analyses the quality of the data gathered by sensors. A set of premises was used to ensure the relevance of the data regarding its quality and usefulness. The last layer tackles uncertainty (e.g., missing values, obsolete, irrelevant, or ambiguous) and removes it from the context analysis. A classification provided by a decision tree was used to grant relevant data to replace the faulty ones. Thus, the framework ensures that the data being used to process context is well-structured, has high quality, and is as complete and updated as possible. In worst-case scenarios, it can, at least, identifying the source of problems. The validation of the proposal was conducted through a series of case studies developed based on two public datasets. Their content refers to ordinary Activity Daily Living (ADL) tasks, and they were built with actual data collected in test environments. Even considering that the datasets have distinct structures, the results evidenced verified an improvement in the quality of the context data. |
|---|---|
| Autores principais: | Freitas, Leandro Oliveira |
| Assunto: | Context awareness identification of uncertainty uncertainty handling Sensibilidade de contexto identificação de incerteza tratamento de incerteza |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | tese de doutoramento |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Ambient Intelligence (AmI) solutions are capable of acting autonomously to benefit human beings. This field of investigation is directly related to other domains, aiming to improve user experience by developing context-aware applications. Context information is naturally dynamic, uncertain, or incomplete. Thus, it is crucial to study ways to handle problems in context data to build applications that run autonomously. The main objective of this Ph.D. thesis is to identify elements that result in the characterization of uncertain contexts. For that, sources and types of uncertainty in intelligent environments were identified. A framework with three layers to validate context data was proposed. The first one defines an attribute grammar to ensure the structure of the data. Restrictions (expressed as logical formulas) associated with the rules of the grammar enforce the system to work with well-defined values. Pieces of context data should have quality, be relevant and be complete to be used by a context-aware system. Thus, the second layer analyses the quality of the data gathered by sensors. A set of premises was used to ensure the relevance of the data regarding its quality and usefulness. The last layer tackles uncertainty (e.g., missing values, obsolete, irrelevant, or ambiguous) and removes it from the context analysis. A classification provided by a decision tree was used to grant relevant data to replace the faulty ones. Thus, the framework ensures that the data being used to process context is well-structured, has high quality, and is as complete and updated as possible. In worst-case scenarios, it can, at least, identifying the source of problems. The validation of the proposal was conducted through a series of case studies developed based on two public datasets. Their content refers to ordinary Activity Daily Living (ADL) tasks, and they were built with actual data collected in test environments. Even considering that the datasets have distinct structures, the results evidenced verified an improvement in the quality of the context data. |
|---|