Prediction of student performance in the academic field creates significant challenges in developing reliable and accurate diagnosis models. Through the use of online learning behavior data, this paper may assist teachers in identifying students with learning challenges in advance and providing timely assistance. A novel technique called Tukey Regressive Hoover indexed Deep Shift Invariant Structure Neural Network (TRHIDSISNN) Model is introduced for student behaviour analysis with lesser time consumption. Initially, the student data and features are collected and transmitted to the input layer. After that, the features of collected student data are analyzed in hidden layer 1 with help of the Tukey Regression. The correlation between one or more independent features is identified to find the dependent feature. The relevant features are sent to the hidden layer 2. In that layer, the Hoover index is applied for analyzing the training and testing features. Finally, the hidden layer result is sent to the output layer where the hyperbolic tangent activation function is used to classify the data that belongs to that particular class. Based on the classification, the student grade level is predicted as high, medium and low based on their behavior gets displayed. Experimental assessment is carried out using different parameters such as prediction accuracy, false-positive rate, prediction time, and space complexity with respect to the number of student data.  The discussed results show that when compared to state-of-the-art approaches, the suggested TRHIDSISNN model achieves higher accuracy with shorter prediction times.