Cluster Analysis to Estimate the Difficulty of Programming Problems

Abstract

Programming is one of the vital skills for the next generation. Currently, there are many online platforms where programmers compete and solve programming problems. Those platforms are composed of problems with varying degree of difficulties. For expert programmers, the difficulty level is not a concern, but it is very important for novice programmers to approach programming problems based on their experience and level. Thus it is important to construct an expert system which can categorize the programming problems based on their difficulties. In our research, we have proposed an expert system which is based on fuzzy rules derivation. These fuzzy rules have been derived by performing cluster analysis on submission log data of Aizu Online judge database. Different clustering algorithms were examined based on the features of these programming problems. The performance of the expert system was compared with 3 different learning models (Decision tree, Random forest, K-nearest neighbor). A high accuracy score on the testing set proves the validity of our constructed fuzzy rules for the expert system.