Improving The Student Evaluation By Predicting Student Preferences Project Abstract:
While preparing oneself for an examination, it is always helpful if one could get an estimate of the topic-wise toughness in his/her course of study. Such an estimate would enable to plan the strategy of preparations to maximize the performance for an examination. One can decide upon what areas to study more or which require special attention. Also it could help improve student evaluation in an exam based scenario where users would be tested on their online performances. In order to help get such an estimate, we attempt to predict the probability of answering a question correctly given the past performance of students who have underwent similar tests and answered questions on the same topics. This will not only help students figure out what areas they are weak in but also design tests which help better measure what a student actually knows based on his grade hence evaluated. The data in our research comes from students studying for GMAT, SAT, and ACT at an online test portal, www.grockit.com . Starting with few state-of-the-art techniques. We would attempt to predict, for each question attempted in the test set, the probability of student answering the question correctly. We would introduce an online unsupervised learning algorithm based on the paradigms of Dynamic Bayesian Networks.
By student evaluation, we mean classifying students on the basis of their skills/strengths. Now, there are a wide range of areas of education. Strength of a student in one of them may be very impressive while he may be relatively very weak at the same time on the other. Therefore arises a need of a topicwise student evaluation wherein a student gets assigned a grade for every possible independent fields of his education paradigm (e.g. Commerce, Science) just like how the education institutes function.
Exploring the problem:
Not only does student evaluation as mentioned help a student in getting aware of his standards but such a plan could help designing a model which could give a good measure of quality of a student and his topicwise strengths/toughness. Since its evaluations that we are talking about, the model could be extended to the fields of tutoring too. Now, how to go about to achieve the target? How can one
decide upon a student’s class as early as possible with sufficient accuracy as well as in the least time of exposure to student? We do it by predicting a student’s answer on every question instance posed for hims. If we can somehow get an estimate of what is the probability of his answering the question correct or incorrect, then it could help us get a measure of how well equipped he is in the areas related to that question. The more probable of him answering, the better his grade is.
At prior , what approach we wish to conquer is :
Feed a certain number of questions to the user keeping a record on the regulation scoring and predictions of his answers. For quicker evaluation use the concept of related topics in a question. Performance in one would help infering about the others. Using the records mentioned, once we get an estimate of his class in each such topic, we can go on and deal thereon with the results depending upon our
application. E.g. We could stop there and report him qualified or ineligible if the environment was an exam or we could go on making him practice upon his weaker fields if this was a tutoring session.
Download Improving The Student Evaluation By Predicting Student Preferences Project Report