Temporal case-base maintenance

Resumen

Case-Based Reasoning (CBR) is a problem-solving methodology that solves problems by analogy with previously solved problems. The basis of CBR is a case, an independent piece of knowledge that associates the description of a problem with its solution, where cases are retained in a knowledge-source known as a case-base. The amount of cases may be a sign of the expertise of the CBR system for solving the problem domain. However, having a large case-base does not guarantee an improvement of the problem-solving capability. On the contrary, an accumulation of many cases may lengthen the response time of the reasoning process and, in certain scenarios, negatively affect the correct solution of certain types of problem. Case-Base Maintenance (CBM) tasks reduce the number of cases within the case-base without affecting the problem-solving accuracy of the reasoning process. CBM is essential when CBR is used in time dependant domains where CBR has to include temporal representation techniques in case descriptions. Nevertheless, temporal representation implies more complex case structures and makes it more difficult and costly to quantify the similarity between cases. This means the case-base should be as small as possible without harming its problem solving capabilities. However, to our knowledge, no algorithm has been proposed to perform CBM in case-bases with temporal cases. In this thesis, we propose: (i) an evaluation method to study the effects of using CBM algorithms on CBR performance; (ii) a temporal framework for use in temporal CBR; and (iii) a set of temporal CBM algorithms. In addition, a new CBM algorithm based on a multiobjective optimization evolutionary approach is proposed. Lastly, our proposals and hypotheses are tested with data of elderly people monitored at home. In particular, the experiments conducted confirm the suitability of our proposed evaluation method to study the consequences of using CBM. Moreover, the experiments also support our initial hypothesis that it is possible to successfully perform a maintenance task on temporal case-bases with the proposed temporal CBM algorithms.