This paper aims to investigate the important factors that affect the value of the minimum passing level (MPL) of competency achievement and find the best method to predict it. The MPL of competency achievement is the value that represents the minimum passing score of examination related to the competency. Different schools may have a different value of the MPL because the MPL is defined based expert opinion on several uncertainty aspects and conditions at each school. This paper proposes the combination of rough sets and fuzzy signature method to predict the category of the MPL. The rough sets method is applied to reduce unnecessary features for classification and find the important factors to predict the MPL. The fuzzy signature is employed to predict the category of MPL based on the selected features. The method proposed in this paper consists of several stages, namely data collection and pre-processing, features selection, predict the category of the MPL using the combination of rough sets and fuzzy signatures method, and performance evaluation. Fifteen headmasters and sixty teachers of elementary schools participated in the data collection process. Based on the experiment with 203 objects data we achieved 97% accuracy in the prediction of MPL. The proposed method succeeded to identify the important factors on predicting the MPL on the complexity of competency and resource capacity of the school aspect. We obtained the improvement for accuracy of the complexity of competency prediction of 8.5% from the best method in the previous research.

This paper is part of research in developing a competency-based assessment system for mathematics in Indonesian elementary school environment. An essential task is to accurately classify questions based on competency and difficulty level. Thus, an expert system is needed to classify those questions since competency information is often manually defined by experts. The objectives of this work are replacing a human expert’s role in the related knowledge engineering process and providing a rule-based expert system to supersede an expert to classify the questions. Five types of the rule-based algorithm: OneR, RIPPER, PART, FURIA, and J48, were applied to the dataset, which is comprised of 9454 real mathematics examination questions collected from several Indonesian elementary schools. Following the knowledge engineering principles, these algorithms generated the classification rules based on a pattern of the data. The rules of the best performing algorithm were utilized by a knowledge base for inference. Finally, to be able to fully measure the system performance, ten expert teachers were involved in the question classification step. The results confirm that the system meets the stated objectives in classifying the competency and the difficulty level of a question automatically.

Complexity of competency (CoC) expresses the difficulty level of a competency. The CoC is one of the important parameters for determining the minimum passing level of competency in an assessment system. In Indonesia, the value of CoC is defined by experts based on conditions of subject, students and teachers in each school. The definition process is determined subjectively, where different experts may evaluate CoC in different ways. This is a problem of data classification that requires an automated tool that copes with the amount of data and produces uniform results. To apply an intelligent classifier is essential to solve the issue. This study aims to find the best method for classifying the complexity of competency in Elementary School. Four supervised learning techniques, namely, Naïve Bayes, Multilayer Perceptron, Sequential Minimal Optimization, and RIPPER, were implemented to analyze the dataset. Based on an experiment with 203 data, we found that the Multilayer Perceptron achieved the best performance in the sense of Mean Absolute Error, Root Mean Squared Error, and Receiver Operating Characteristic value. At the same time SMO is better than all other methods in precision, recall, and F-Measure.