Performance Evaluation of feature extraction model to identify student appraisals

Abstract: Educational Data mining techniques plays an important role in educational institution. It can be used to understand the difficulties arising in the teaching-learning professions. In machine learning, feature selection or Attribute analysis usually emerges as a pre-processing step. Feature selection is the problem of choosing a small subset of feature that ideally is necessary and sufficient for predictive / decision-making type of learning tasks. This study proposes a framework for identifying the most significant attributes towards academia, for the performance of second year students of computer science and application course. The authors realize that the some features are non-changeable and so do not contribute in upgraded academic performances of the students as they do not reveal any added academic effort. In this study, authors decided to work upon only external attributes of students by assigning weights that reflect their residual efforts put in for those attributes. The model is able to extract the fitness precedence relations of external efforts put up by student belonging to both „above-risk‟ and „at-risk‟ categories in their on-going course. The end-user can make use of these precedence relations to identify and resolve the most unfit governing factor for upgrading students‟ appraisals. The accuracy of these precedence relations is computed upon the most popular feature extraction (FE) algorithm „RELIEF „The model accuracy of 75% indicates the encouraging results in the direction of identifying graded precedence of the participating model attributes.

Keywords: Feature Extraction, Feature Selection, External Factors, Naïve Bayesian classification. Attribute Relevance, Precedence Relations, Relief Algorithm, Nearest-Hit (Z+), Nearest-Miss (Z-)