Development of Effective Artificial Neural Network Model using Sequential Sensitivity Analysis and Randomized Training

As the machine learning algorithms evolve, there is a growing need of how to train the algorithm effectively for the large data with available resources in practically less time. The paper presents an idea of developing an effective model that focuses on the implementation of sequential sensitivity analysis and randomized training approach which can be one solution to this growing need. Many researchers focused on the implementation of sensitivity analysis to eliminate the insignificant features ands reduce the complexity in data selection. These sensitivity analysis methods relatively take a large time for validation through modeling and hence found impractical for large data. On the other hand, the randomized training approach was found to be the most popular approach for training the data but there is a very brief explanation available in research articles on how this training method is meaningful in getting higher accuracy. The current work focuses on the use of sequential sensitivity analysis and randomized training in an artificial neural network (ANN) for high dimensionality thermal power plant data. The sequential sensitivity analysis (SSA) technique includes the use of correlation analysis (CA), Analysis of variance (ANOVA), Akaike information criterion (AIC) in a sequential manner to reduce the validation time for all possible feature combinations. Only selected combinations are then tested against different training methods such as downward extrapolation, upward extrapolation, interpolation and randomized training in ANN. The paper also focuses on suggesting the significance of training with randomized training with comparison-based qualitative reasoning. The statistical parameters, mean square error (RMSE), Mean absolute relative difference (MARD) and R Square (R^2)were accessed for validation purposes. The research work mainly useful in the field of Ecommerce, Finance, industry and in facilities where large data is generated.