Klasifikasi Multilabel Pada Gaya Belajar Siswa Sekolah Dasar Menggunakan Algoritma Machine Learning

Abstract

Learning style plays a very important role in determining the success of a person's learning process. An individual generally has a combination of all three existing learning styles including Visual, Auditorial, and Kinesthetic. However, what distinguishes the abilities of individuals from each other is how the dominant combination of each learning style is or not, so it is important to identify. This study aims to build a multi-label classification model to classify the learning styles of elementary school students. The machine learning algorithms used to build the model are Decision Tree, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP). The performance of these four models is compared using the Hamming Loss, Accuracy, Precision, Recall, and F1-Score performance metrics. The Classifier Chains method is implemented to provide capabilities to KNNs and SVMs that cannot directly handle multi-label classification problems. The dataset used in this study is the Data Set of Learning Style Preference. The separation of the dataset was made into three different forms of data sizes, including: Data I: 90% training, 10% testing; Data II: 80% training 20% testing; and Data III: 70% training 30% testing. Cross-validation using K-Fold Cross Validation with a k-value of 10-fold was also applied to the training data. Based on testing, the best performance was obtained on the Decision Tree model with a hamming loss of 0.014, which indicates a very low prediction error rate per individual label.  A recall value of 99% indicates that the model is able to detect almost all positive labels correctly, and an F1-score of 98% indicates that the model has excellent and balanced performance, without bias against both positive and negative label predictions. The performance of the Decision Tree model was followed by MLP, SVM, and KNN which showed lower results.