Time Series Data Transformation to Measure the Complexity of a Learning System and To Predict the Performance of a Learner with Data Mining Techniques

Abstract

Advancement in technology has made everything online which results in tremendous accumulation of data in various formats. Every user s information becomes data that is stored as records in a huge database. Data mining techniques are employed to discover useful knowledge from this large database. The education system nowadays employs learning analytics to predict the outcome of students and teachers every year. This makes them take improvement decision that enhances the performance of both students and teacher. Therefore a large database is maintained over which Educational Data Mining techniques are applied to extract useful information. The educational dataset of this research is composed of student s action recorded during laboratory sessions. The action includes mouse movements, still cursor, clicking, highlighting, idle time and lots more. This dataset records the action as a time-series data. Time-series data is a non-linear, multivariate dataset that is huge in size as well as more complex. They are unstructured dataset with comma separated values, which is very challenging to apply DM techniques. To reduce this complexity, this research work proposes a conversion algorithm, in which text-based log time series data is transformed into numerical data. Using this algorithm Educational Process Mining (EPM) and Educational Data Mining (EDM) tasks can be performed effectively with the given dataset using clustering, classification, and association rule mining algorithm. The proposed conversion algorithm is implemented in MATLAB7. The proposed tree-based EDM model is applied with conventional data mining metrics, Neural network Algorithms namely multi-layer Perceptron algorithm and radial basis function, and then with association rule mining algorithm. The performance of the students is accessed through various metrics namely precision, accuracy, recall, f_score, confidence, and support obtained using the above-mentioned implementations. The results are compared with Mccabe s Cyclomatic Complexity (CM).