PMSIEMDL: Design of a Pattern analysis Model for identification of Student Inclination towards different Educational-fields via Multimodal Deep Learning Fusions

Abstract

Identification of student inclination towards different educational fields requires integration of deep pattern learning models with temporal data analysis techniques. These techniques are highly context sensitive, and cannot be scaled for analysis of students that have an interest in multiple domains. Moreover, existing deep learning models are highly complex, and showcase moderate performance when used on real-time datasets. To overcome these limitations, this text proposes design of a Pattern analysis Model for identification of Student Inclination towards different Educational-fields via Multimodal Deep Learning fusions. The proposed model initially collects data samples from a large number of students, and segregates them into different classes. These include social data, personal habits data, education data, family related data, performance data and future inspiration data classes. These datasets were combined with a customized psychological questionnaire which was curated by experts in the field of student counselling & psychology. Based on student responses, their entity specific classes were generated, that were separately trained via different Convolutional Neural Network (CNN) Models, which assists in identification of student-performance at individual-class levels. These performances are compared with existing inclination datasets via a fusion of Long-Short-Term Memory (LSTM) & Gated Recurrent Neural Network (GRNN), which assists in identification of correlation between subject-level inclinations & their entity classes. This provides with a probabilistic map of different subjects towards which the student might be inclined, and assists them to select their study streams. The generated map was validated for multiple students, and recommendations were made based on higher probability values, which assisted in identification of student inclination levels. The model was evaluated under large datasets and its performance was compared with various state-of-the-art methods under different scenarios. Based on this comparison, it was observed that proposed model was capable of achieving 8.5% better recommendation accuracy, 4.9% higher prediction precision, 6.5% better recommendation recall & 2.9% better Area Under the Curve (AUC) levels, which makes it highly useful for a wide variety of student inclination use cases