Improvements of fuzzy C-means clustering performance using particle swarm optimization on student grouping based on learning activity in a digital learning media

The field of learning media has been developing rapidly in recent years, especially in an effort to support students' learning process. The amount of recorded learning process data has also significantly increased. The recorded data represents the students' thinking process in building a solution for a problem. The sheer size of the recorded data proves to be quite a challenge in an effort to mine the students' thinking process, especially when done manually. Additionally, to group the recorded data into clusters is also another form of challenge that needs to be faced. In general, the entire process of mining students' thinking patterns aims to utilize the data to gather hidden information which can also be used to give appropriate and proper feedback to the students. This paper aims to employ the Fuzzy C-Means and Particle Swarm Optimization (FCMPSO) method to cluster students based on their learning activity to a digital learning media and compare its performance to original Fuzzy C-Means (FCM) method. Particle Swarm Optimization (PSO) algorithm is proposed to optimize the performance of the FCM algorithm, in which this algorithm is inherently sensitive towards centroid on the initial clustering process that utilizes the Silhouette coefficient as an evaluation method. Based on the experiments that have been done to 12 assignments, each assignment forms a different number of optimal clusters. This shows that each student faces and uses different strategies to solve their assignments. The formed groups are dominated by two major clusters, namely the high-performance students, and the low-performance students. Additionally, the adaptation of PSO to FCM improves the clustering quality significantly based on the observed average Silhouette coefficient.