Mining educational data to predict students performance

Abstract

Information is everywhere in a hidden and scattered way. It becomes useful when we apply Data mining to extracts the hidden, meaningful, and potentially useful patterns from these vast data resources. Educational data mining ensures a quality education by analyzing educational data based on various aspects. In this paper, we have analyzed the academic results and behavior of some engineering students. For this study, we collect data from 80 students from the CSE department. We gather data from mark sheets and other relevant factors that accelerate the results, collected through a survey. Our main goal is to predict the students’ performance. According to this prediction, the counseling department will guide them in advance so that those who are likely to have bad results can do better. The classification can be based on various aspects, as many factors improve the educational system. We have created two datasets focusing on two different angles. Our first dataset classifies and predicts the category of a student (good, bad, medium) on a specific course based on their prerequisite course performance. We have implemented this in the artificial intelligence course. Our second dataset also classifies and predicts the final grade (A, B, C) of any random subject, here we organize our data such a way where it will only focus on how their performance was till the midterm exam. We analyze and compare six classification algorithms. We have focused on all aspects of an algorithm, not only the accuracy level but also the complexity and cost. We have built two final models for two of our datasets based on a decision tree and the naive Bayes algorithms accordingly.

Appendix

• ANN: ANN or Artificial Neural Network is a computational algorithm. This algorithm has been designed by using the concept of human neuron. It process information as like human brain analyze and processes information. It has the capability of self-learning that enables it to produce better results.

• KNN: KNN stands for K-Nearest Neighbour. It is a simple Machine Learning algorithm based on Supervised Learning technique. This algorithm find outs the similarity between new and existing data and puts the new data into the most category which is most similar with the available categories.

• Weka: An open source machine learning software with a collection of machine learning algorithms and data preprocessing tools. By using this software users can try out existing machine learning methods on their datasets in a flexible way.