Predicting Reliability of Response in Online Surveys

Abstract

The use of online surveys is exponentially increasing day by day. From being less time consuming, to being easy to use, online surveys have become highly advantageous. However, the unreliability of participants’ response has become a growing concern. A tool has been implemented, in the attempt of producing a detection mechanism to eliminate unreliable users’ responses on merely the basis of their behaviour, while filling the online surveys such as time taken to answer the questions, clicks, excessive clicking, longer inactivity, changes on already given answers, time taken to answer open ended questions, changes on the screen and activation and changes of form elements like radio buttons, checkboxes and drop downs. All the attributes considered are influential to an extent. Total answer clicks, number of responded questions, checkbox changes during the survey and the status of idle are the most influential attributes in identifying the reliable response of online surveys. The algorithms used, namely, Naïve Bayes, Logistic Regression, Decision Tree, Support Vector Machine and Random Forest are also quite reasonable considering that the accuracy for all of them were above 50%. The most influential algorithm was Logistic Regression.