<span lang="EN-US">The fuzzy time series (FTS) is a forecasting model based on linguistic values. This forecasting method was developed in recent years after the existing ones were insufficiently accurate. Furthermore, this research modified the accuracy of existing methods for determining and the partitioning universe of discourse, fuzzy logic relationship (FLR), and variation historical data using intervals ratio, cross association relationship, and rubber production Indonesia data, respectively. The modified steps start with the intervals ratio to partition the determined universe discourse. Then the triangular fuzzy sets were built, allowing fuzzification. After this, the FLR are built based on the cross-association relationship, leading to defuzzification. The average forecasting error rate (AFER) was used to compare the modified results and the existing methods. Additionally, the simulations were conducted using rubber production Indonesia data from 2000-2020. With an AFER result of 4.77%&lt;10%, the modification accuracy has a smaller error than previous methods, indicating very good forecasting criteria. In addition, the coefficient values of </span><em><span lang="EN-US">D₁</span></em><span lang="EN-US"> and </span><em><span lang="EN-US">D₂</span></em><span lang="EN-US"> were automatically obtained from the intervals ratio algorithm. The future works modified the partitioning of the universe of discourse using frequency density to eliminate unused partition intervals.</span>

Fuzzy time series (FTS) firstly introduced by Song and Chissom has been developed to forecast such as enrollment data, stock index, air pollution, etc. In forecasting FTS data several authors define universe of discourse using coefficient values with any integer or real number as a substitute. This study focuses on interval variation in order to get better evaluation. Coefficient values analyzed and compared in unequal partition intervals and equal partition intervals with base and triangular fuzzy membership functions applied in two factors high-order. The study implemented in the Shen-hu stock index data. The models evaluated by average forecasting error rate (AFER) and compared with existing methods. AFER value 0.28% for Shen-hu stock index daily data. Based on the result, this research can be used as a reference to determine the better interval and degree membership value in the fuzzy time series.