Novel method for predicting academic performance of students by using modified particle swarm optimization (PSO)

Abstract

There are numerous methods for extracting useful information from data. This paper describes a method for predicting performance of students. This method modifies the basic particle swarm optimization (PSO) algorithm using a set of rules. An attribute is selected from a set of performance attributes of the students. This attribute is used to frame rules. These rules determine the value of a modifying factor. This factor changes the mathematical expression of the function used in PSO for finding the solution. These rules are based on number of students in a particular shift. Other attributes are assigned different indexes. These indexes indicate number of students deviating from average value. The modified PSO algorithm takes the values of these indexes as inputs and generates a solution set which minimizes the values of indexes. A comparison of the solution set given by modified PSO and the solution set with unmodified PSO is presented. A brief outline of the modified PSO is given. The selection of the modifying factor and design of rules is described. These rules are based on the number of students in a particular shift. The different possible classes for the shift attribute are given. Thus, a decision strategy for predicting performance is described.