Moving Studentized Expectile Convolution Connectionist AI for Accurate Employability Prediction in Higher Educational Institutions

Abstract

Employability prediction has received large attention to recognize key predictors of employability among students in higher educational institutions. The student employability prediction is a major concern for higher educational institutions as it takes timely actions for increasing the institutional placement ratio. Different data mining techniques such as classification are used in existing research for predicting the employability of students. But, the existing techniques failed to address the overfitting issues and time consumption remained unaddressed. Therefore, a new convolution based AI model is introduced for performing accurate employability prediction. In this paper, a new Moving Studentized Expectile Convolution Connectionist AI (MStECCAI) Model is introduced to accurately predict the employability with minimum time consumption. The proposed MStECCAI Model comprises four processes namely student information collection, pre-processing, feature selection and classification. During student information acquisition phase, student samples (i.e., student class room videos with audio files, facial expression and head pose movement) are collected from input dataset. Subsequently, the proposed MStECCAI Model utilizes Connectionist AI model with DenseNet Convolutional Deep Belief Network for employability prediction. DenseNet Convolutional Deep Belief Network includes three Restricted Boltzmann Machines (RBMs) layers with hidden layer and pooling layer. First, the number of student samples is given to the visible layer of the DenseNet Convolutional Deep Belief Network. Subsequently, data preprocessing is performed in RBM layer 1 to handle missing data, to detect outliers and to normalize inputs for accurate employability prediction. Followed by, the feature selection process is carried out in RBM layer 2 by using regression analysis. Then, the proposed MStECCAI Model performs the classification process with relevant features in RBM layer 3 with higher employability prediction accuracy. At final stage, the fine tuning process is carried out in proposed MStECCAI Model to reduce the prediction error using nature-inspired swallow search optimization. Finally, accurate employability prediction results are obtained at the output layer. Experimental analysis of proposed MStECCAI Model is conducted with different metrics such as student employability prediction accuracy, precision, recall, F1-score, Matthews Correlation Coefficient, coefficient of determination and student employability prediction time. The quantitatively analyzed results reveal that the proposed model attains higher accuracy in student employability prediction with lesser time consumption when compared to traditional deep learning methods.