Design and Performance Analysis of Gradient-Based Optimization Techniques for Household Power Consumption Prediction

The use of computational intelligence techniques in solving engineering problems is on the rise, demonstrating effectiveness and accuracy in such applications. This paper presents a design and performance analysis of Artificial Neural Networks (ANNs) trained with the Scaled Conjugate Gradient (SCG) and Stochastic Gradient Descent (SGD) algorithms for predicting household power consumption. The results indicate that, while both algorithms were able to satisfactorily model the problem at hand, the SCG optimizer outperforms the SGD optimizer, achieving a Root Mean Square Error (RMSE) of 0.025 and a consistent R-squared value of 0.9996, compared to the best-performing SGD model which recorded an RMSE of 0.95. This study expands existing knowledge on optimization techniques for ANN-based regression and time-series models, providing insights for selecting the most effective optimization strategy for various real-world applications.