KERNEL DISCRIMINANT OPTIMIZED DEEP CONVOLUTIONAL NEURAL LEARNING CLASSIFIER FOR PREDICTIVE ANALYTICS

Abstract

 Data mining (DM) is an emerging field that focuses on applying big datasets of mining techniques to enhance the results and learning process. It involves find out the patterns and correlations within huge sets of data with the aim of forecasting future outcomes. Predicting future outcomes for large datasets is a critical application of data mining. By analyzing historical admission data and other relevant factors, existing predictive models have been developed to forecast the likelihood of various outcomes. However, these models faced challenges when applied to large datasets, affecting their predictive effectiveness. To improve prediction accuracy, a novel model called Kernel Discriminant Optimized Deep Convolutional Neural Learning Classifier (KDODCNLC) is introduced. The primary aim of the KDODCNLC model is to achieve higher prediction accuracy and reduced time consumption. In this model, data is first collected during the data acquisition phase for classification purposes. The proposed DCNN comprises the different layers. At first, data is fed into input layer from Student Performance Prediction dataset. The input data then moves to hidden layer 1, where preprocessing is performed to remove unwanted data. The preprocessed data is passed to hidden layer 2. In this max-pooling layer, feature selection is conducted to identify significant features while discarding others. Then, it is sent to a fully connected layer for classification to reduce the features. In this layer, data is determined by a kernelized discriminant function to provide prediction outcomes. To minimize classification error, hyperparameter optimization in Convolutional Neural Networks (CNNs) is essential, involving fine-tuning of network parameters. The KDODCNLC model employs the Rain-fall optimization algorithm for hyperparameter optimization, aiming to reduce error and enhance prediction accuracy. To conclude, the accurate prediction outcomes are attained at output layer, enabling predictions through less time consumption. Experimental evaluation obtains the different metrics. Outcome quantitatively demonstrates KDODCNLC model enhances the prediction accuracy by 5%, precision by 5%, and recall by 3% and F1-Scoreby 4%, then the conventional methods. Additionally, KDODCNLC achieves a 27% reduction in prediction time than the existing methods.