Accuracy evaluation of prediction using supervised learning techniques

The term Big data is used to refer the huge volume of complex and growing data generated from many distinct electronic gadgets. In the case of Big Data, the commonly used programming methods are not adequate to collect, store and analyze the data within a short period. Statistics as well as machine learning techniques are used for finding patterns and information from large data through data driven decision-making. Big Data analytics gives competitive opportunities in designing business plans for Business Analytics. For analytical purpose, traditionally we use Multiple Linear Regression (MLR) model in the statistical method, a type of Supervised Machine Learning Algorithm. We implemented Cross-Validation Resampling technique with MLR model. The performance of new MLR-Leave-One-Out (MLR-LOOCV) model evaluated using partitioning the whole data set. This technique used to validate the model developed from training data with test data to control the problem like over fitting. The accuracy of such prediction model is very poor. So we propose to build a Multilayer Perceptron Neural Network (MPNN) model with gradient descent learning method to improve the efficiency of prediction model. The new proposed model, MPNN with GD shows accuracy much greater than normal MLR. The data set from UCI machine learning repository is used for simulation methods to check the performance.