Student Performance Prediction Using Convolutional Neural Networks
B.Sai Kalyani1, D.Harisha2, V.RamyaKrishna3, M.Suneetha4
Velagapudi Ramakrishna Siddhartha Engineering College, Vijayawada, AP, India
Abstract: The performance of the students is an important factor for most of the educational institutions. The performance is based on various factors that plays an important factor in the career of student. In the recent years predicting the performance of students has become an important problem in all the institutions.
Modern day educational institutions have adopted continuous evaluation to improve the performance.
In the recent years Neural Networks is being used for predictions, which provides better results when compared to classifiers. The data that is used consists of the number of hours the student has spent on studying the details, his involvement in the activities and other contributing factors as these play a keen role in affecting the performance of the students.
Thus Neural Networks play a key role in predicting the performance. This paper uses Convolutional Neural Networks to predict the performance.
Keywords: Academic Performance, Neural Networks, predictions.
The major issue for the educational systems is to maintain the success rate of the students by helping the students to perform well in the examinations. There are many factors for the institutions to predict the performance of students. One important factor is to determine whether a particular student is eligible to get a scholarship from the institution. The second most important reason to is to identify which students will fail in the semester end so that necessary steps can be taken to improve the performance. Normally the performance of the students is obtained manually by the teacher by observing the past scores or his performance in the classroom. This can be used even for the web based learning because the professor cannot get direct interaction with the student then it becomes difficult to measure the performance. But these assumptions may not work in all cases. As a result various algorithms are used to obtain accurate results.
Initially various data mining classification models such as Na?ve Bayes, Support Vector Machine, Decision Trees can be used for predictions. are used but later it is found that the results are not perfect. Then machine learning algorithms have come into existence where the accuracy percentage is even more better when compared to data mining algorithms. But with the introduction of Neural Networks the accuracy rate has increased rapidly. Neural Networks is a very powerful mechanism for predictions. Which predicts based on the loss calculated for every combination of the values and provides the best prediction. The most important problem is the large amount of data. These records must be transformed, calculate the percentage and provide the accurate predictions. But Neural Networks puts an end to this problem thereby enabling its usage for large amounts of data.
Ayusha Ashraf, Sajid Anwer, Muhammad Gufran Khan  compared various data mining techniques, classification algorithms to obtain the result. The main attributes that affect the students performance are also identified. Elbadrawyetal  used ANSI models by normalizing the results obtained in previous results. Three ANFIS models used models: ANFI GaussMF,ANFIS-TriMF, and ANFIS-GbellMF models and ANFIS-GbellMF model showed the best result. S. Kotsiantis, et al  used six different machine learning algorithms to identify the poor performer students who are performing distance education. T.Prabha, D.Shanmuga Priyaa  used an evolutionary method that used Artificial Neural Networks and reducing the poor performnance of the ANN using swarm optimization so that the weight values are properly updated thereby increasing the accuracy. Bendangnuksung, Dr.Prabu  used Deep Neural Networks to categorize the students based on their performance which observed a better accuracy than other models.
Convolutional Neural Networks (CNN) helps in designing the system to simulate the way, how the human brain analyzes and processes the given information. This technology helps in solving the problems that are impossible to solve by human brains and that beyond the imagination. The steps used are:
Step 1: Sequential is used to initialize the neural network.
Step 2: The first CNN layer uses ReLu activation function and output is given to next layer with same activation function but with dense value 32.
Step 3: These two layers are repeated. Sparse_categorical_entropy is calculated and the weight values are updated.
Step 4: The output is provided to the next layer with activation function sigmoid and accuracy value is obtained.
The detailed architecture is shown in Fig 3.1.
14868942733482-556591620299-556591620299-556592733482148686616199683323645161234844686337534413323618363993139145136399320777202473325Predicting the result
1946412411205Obtaining loss accuracy
1919911335129Train the model
20281901334135Implement the model
37768701278392Tensor flow for holding labels and features
00Tensor flow for holding labels and features
377802986774Dividing into test and train data
1900362109551Encoding the dependent variables
0Encoding the dependent variables
9541661843Loading the dataset
Step 1: The dataset is in the form of .csv and is read to a variable.
Step 2: The data is extracted and is shown in fig 4.1
Step 3: The labels are converted using Label Encoder shown in Fig 4.2
Step 4: The CNN model is implemented with 3 dense layers and activation function .Also the loss value is calculate and the values are updated as shown in Figure 4.3
Step 5: The dataset is divided into train and test. The model is fit and 100 epochs are trained with batch_size 5 as in Fig 4.4.
The details of a random student are provided as input by the user as shown in Fig 4.5 and the percentage of performance is predicted.
The input is processed and the performance of a random student is as shown in Fig 4.6.
ConclusionThe main purpose is to develop model that predicts the performance of students. The study also shows that it is possible to predict the student graduation performance, which is based on the major impacting factors that affects the final output of the student. The experiment results are done in Python and convolutional neural networks is used for predicting the performance. It is concluded that the meta-analysis on predicting student’s academic performance motivated us to do further research work in our own educational environment.
 Ayusha Ashraf, Sajid Anwer, Muhammad Gufran Khan.: A comparative study of predicting students performance using data mining techniques , ASRJETS, 2018.
 Asmaa Elbadrawy, Agoritsa Polyzou, Zhiyun Ren, Mackenzie George Karypis, Huzefa Rangwala: Predicting Student Performance using Personalized Analytics , IEEE Computer Society, April 2016.
 S. Kotsiantis, et al. (2003): Preventing student dropout in distance learning systems using machine learning techniques Applied Artificial Intelligence , 18(5), pp.411- 426.
 T.Prabha, D.Shanmugha Priyaa An Evolutionary Approach on Students Performance Prediction and Classification,ijpam.eu,2018.
 Bendangnuksung, Dr.Prabu Students Performance Prediction Using Deep Neural Network,International Journal of Applied Engineering Research, 2018.
 Abimbola R.Iyanda, Olufemi D.Ninan, Anuoluwapo O.Ajayi, Ogochukwu G. Anyabolu Predicting Student Academic Performance in Computer Science Courses:A Comparision of Neural Network Models,MECS, 2018.
 C.Jayasree, K.K.Baseer Predicitng Student Performance to Improve their Employability by Applying Data Mining and Machine Learning Techniques, JCSE,2018.