Ensemble clustering and feature weighting in time series data

Ensemble clustering is an important approach in machine learning, which combines multiple hypotheses to minimize the risk of selecting a wrong hypothesis or local minimum. In this study, two parallel and distributed frameworks for ensemble clustering and time series data prediction are presented. The second framework is presented with a higher distribution level than the first framework. Both frameworks were implemented using the MapReduce programming model, with the Relief algorithm's various versions used for feature weighting, including Multisurf*, ReliefF, Simba-Sc, and I-Relief. Additionally, a new version of Relief called M-Relief is introduced and compared to other versions. To analyze the proposed frameworks' performance, Irish weather data, energy consumption data from PJM, and Spanish weather data from the Kaggle dataset were selected. The study's results demonstrated higher clustering accuracy and diversity compared to the basic Relief, with performance measured in terms of memory, runtime, and error metrics.