Fuzzy Time Series K-Medoids for Inflation Forecasting in West Sulawesi Province

Abstrak

Inflation is a key indicator of macroeconomic that reflects the price stability of a region and directly influences purchasing power, business planning, and public policy. West Sulawesi, an agriculture- and fishery-dependent province, exhibits a highly fluctuating monthly inflation pattern that is difficult to capture with conventional linear forecasting models. This study proposes a Fuzzy Time Series (FTS) model combined with K-Medoids clustering to forecast monthly inflation in West Sulawesi Province using Statistics Indonesia (BPS) data from January 2019 to December 2024 (72 observations). K-Medoids clustering, initialized using the optimal number of clusters selected via the Pseudo-F statistic and validated with the Silhouette Coefficient, was used to partition the historical data into representative intervals prior to fuzzification. Fuzzy Logical Relationships (FLR) and Fuzzy Logical Relationship Groups (FLRG) were then constructed to model the transition pattern between fuzzy states, and defuzzification was performed to generate numerical forecasts. Model accuracy was evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE). The optimal number of clusters was found to be k = 7, supported by the highest Pseudo-F value (325.28) and Silhouette Coefficient (0.574). The FTS K-Medoids model achieved an MAE of 0.33, an MSE of 0.18, and a MAPE of 1.45%, corresponding to a forecasting accuracy of 98.55%, which falls into the “very good” category. The out-of-sample forecast for January 2025 was 0.253%, indicating a stable and controlled inflation outlook. These results demonstrate that the FTS K-Medoids approach is an effective, distribution-free alternative for forecasting nonlinear and volatile regional inflation data