<![CDATA[Missing values in financial time series data can affect the information structure of the data and impact the clustering results obtained. This research aims to evaluate the performance of several time series data imputation methods on the quality of categorical time series clustering on financial sector stock data on the Indonesia Stock Exchange. The imputation methods compared include linear interpolation, spline interpolation, and Kalman smoothing. The research data is in the form of daily closing prices of 92 financial sector stocks for the period 2 January 2023 to 31 October 2025. Numerical clustering was carried out using K-Means Time Series based on Dynamic Time Warping (DTW), while categorical clustering was carried out using K-Medoids with the Gower distance measure in two categorization schemes, namely five and seven categories. Evaluation of suitability between numerical and categorical clustering was carried out using the Rand Index (RI), Fowlkes–Mallows Index (FMI), and Jaccard Index. The research results show that the imputation method produces different clustering qualities. Linear interpolation provides the best and most consistent performance compared to other methods, especially in the seven-category scheme with an RI value of 0.6417, FMI of 0.4338, and Jaccard Index of 0.3256. These results show that linear interpolation is better able to maintain the information structure of the data in the categorical time series clustering process compared to spline interpolation or Kalman smoothing.]]>
Copyrights © 2026
