Predictive Machine Learning Approach for Complex Problem Solving Process Data Mining

Problem-solving is considered to be an essential everyday skill, in professional as well as in personal situations. In this paper, we investigate whether a predictive model for a problem-solving process based on data mining techniques can be derived from raw log-files recorded by a computer-based assessment system. Modem informatics-based education relies on electronic assessment systems for evaluating knowledge and skills. OECD's PISA 2012 computer-based assessment database was used, which contains a rich problem-solving dataset. The dataset consists of detailed action logs and results for several problem-solving tasks. Two feature sets were extracted from the selected PISA 2012 Climate Control problem solving task: a set of time-based features and a set of features indicating the employment of the VOTAT problem-solving strategy. We evaluated both feature sets with six machine learning algorithms in order to predict the outcome of the problem-solving process, compared their performance and analyzed which algorithms yield better results with respect to the observed feature set. The approach presented in this paper can be used as a potential tool for better understanding of problem-solving patterns, and also for implementing interactive e-learning systems for training problem solving skills.

Automated summary

This study investigates the potential of deriving a predictive model for a problem-solving process from raw log-files, using data from OECD's PISA 2012 computer-based assessment database. Two feature sets were extracted from the dataset and evaluated with machine learning algorithms, aiming to understand problem-solving patterns and improve e-learning systems for training such skills.