Numerous Fuzzy Time Series (FTS) models have been proposed in scientific literature during the past decades or so. Among the most accurate FTS models found in literature are the high order models. However, three fundamental issues need to be resolved with regards to the high order models. First, current prediction methods have not been able to provide satisfactory accuracy rates for defuzzified outputs (forecasts). Second, data becomes underutilized as the order increases. Third, forecast accuracy is sensitive to selected interval partitions. To cope with these issues, a new high order FTS model is proposed in this thesis. The proposed model utilizes aggregation and particle swarm optimization (PSO) to reduce the mismatch between forecasts and actuals. Comparative experiments confirm the proposed model's ability to provide higher accuracy rates than the current results reported in the literature. Moreover, the utilization of aggregation and PSO, to individually tune forecast rules, ensures consistency between defuzzified outputs and actual outputs, regardless of selected interval partitions. As a consequence of employing these techniques, data utilization is improved by: (1) minimizing the loss of forecast rules; (2) minimizing the number of pattern combinations to be matched with future time series data. Finally, a fuzzification algorithm, based on the trapezoid fuzzification approach, has been developed as a byproduct. The proposed algorithm objectively partitions the universe of discourse into intervals without requiring any user defined parameters.