Artificial Intelligence

The proposed recommendations  system involves:

  1. Authentication
  2. The computation of score (RNN)
  3. Clustering
  4. Prediction of performance
  5. Provides recommendations 


  1. The authentication weight will be estimated from identity id, password, PIN number, and last test attended data.
  2. The student will be authenticated each time while attending their online tests. From these security credentials, a weight value will be estimated using which the student is authenticated.
  3. Once they complete their tests online, then their score values will be computed for determining his/her performance.

Student Score Estimation

  1. The overall scores of individual students are estimated from Recurrent Neural Network (RNN) which is composed of preprocessing, feature extraction and finally score estimation.
  2. Firstly, pre-processing is performed with the processes of data cleaning and elimination of data duplication. On elimination of duplicate records from the dataset, the attributes will be extracted for score estimation.
  3. The features from student engagement are class attendance, number of clicks, course concept and preferred incentive. Then the score features that are extracted are multiple choice questions, writing and final examination results.
  4. Based on the two major constraints as the student engagement and student score the student’s performance is evaluated.
  5. The student with better engagement and score will have higher performance. The good scoring student will certainly have better engagement in the class, if not then their incentives will be very low.


  1. After determining the student score values, they will be clustered into three groups as poor, average and excellent.
  2. Density-based spatial clustering algorithm (DBSCAN) using Mahalanobis distance clustering will be employed to cluster students based on their performances in terms of scores.
  3. The top scored students are excellent, similarly based on their score values they are clustered using Mahalanobis distance in DBSCAN.
  4. This clustering algorithm is preferred:
  5. It is capable to deal with a larger sized data.
  6. The distance measurement reflects on the clustering and hence Mahalanobis distance is used which is suitable for similarity measurement.
  7. After the clusters are constructed, cluster will be validated the strength by measuring purity and entropy of the clusters.
    1. If the entropy and purity is higher than the pre-defined threshold value then the student’s performance is predicted from the constructed cluster.
    1. If not, then the clusters are re-created by estimating distance for the data points.
  8. In clustering, the students will be grouped within certain ranges which cannot identify one’s performance.

Student Performance Prediction

  1. The cluster values will be taken as input and the student’s results are estimated.
  2. In order to identify the performance of individual students, a threshold-based map-reduce (TMR) process will be applied for categorizing an individual’s performance.
  3. The exact performance of each student is predicted from the clusters that groups the students based on the score values.
  4. According to the performance of the students, the threshold is predicted and then they are reduced.
  5. In this TMR, the threshold value is defined in order to predict student’s performance. The value of threshold is determined based on the student score.
  6. Map: Initially, the input cluster is split into individual score values of the individual student. Then the map phase is executed as the first phase. In this phase, each split value from the cluster is operated based on the mapping function
  7. Reduce: In shuffling the mapped output values are processed, by consolidating the matched records in mapping phase. This shuffling enables to remove duplicate values that exist and then it groups the similar values
  8. This process ensures to predict the accurate performance of the individual student.

Student Recommendations

  1. The recommendations for the student will be provided using State–action–reward–state–action (R-SARSA) which is a reinforcement learning algorithm.
  2. A set of rules will be determined from the mean value of student scores and engagement.
  3. Along with the probability value of the rule, the current performance will be also taken into account for recommendations.
  4. The states in R-SARSA will be defined based on the probability value and current performance.
  5. The recommendation for the student is presented to improve the learning efficiency of the student.

The proposed system is modeled for predicting student performance and deliver recommendations for average and poor learners.

  1. AISAR system will be developed in Hadoop environment with Java programming language.
  2. Next student registers or logins with ID, Password, PIN Number, and Last Attended Test Date.
  3. Performing the proposed concept.
  4. Recurrent Neural Network will be implemented to calculate the individual score by considering the Engagement and Examination score.
  5. Implementing Data Clustering with DBSCAN (Density based spatial clustering algorithm) using Mahalanobis distance clustering:
  1. Validate the cluster construction by considering the ‘Purity’ and ‘Entropy’.
  2. Next, the score of the students is mapped using TMR (Threshold MapReduce):
  3. Finally, provide the recommendation for the students using reinforcement Learning (R-SARSA).