Real-time Computation of Features Based on Probabilistic Methods in Machine Learning

Abstract

Traditional energy sources at a residential level for a sustainable environment. Energy consumption in Kilowatts(kW) in each household varies depending on the appliances used. To store energy in an emergency and maintain the continuity of supply-demand energy predictions have been developed. General Regression Neural Network(GRNN) uses probabilistic machine learning with uncertainty into consideration for prediction results. GRNN implemented on the energy consumption dataset containing humidity and temperature data columns considers errors and value uncertainty of the data. The GRNN is implemented against commonly used Tree-based algorithms like  Random Forest Regressor, XGBoost Regressor, and Extra Tree Regressor performs well. The GRNN slightly outperforms the tree-based regressors with an R2(accuracy scale for neural networks)score of 0.61 and the best tree regressor of 0.58. The probabilistic approaches have better accuracy than the non-probabilistic approaches for prediction problems as they use distribution as inputs rather than valued data. Advanced data acquisition techniques can improve the R2 scores of GRNN by providing more data for the neural network model.