Explainable Artificial Intelligence for Definition of Inputs in Neural Networks and K-nearest Neighbors Forecasting of Electricity Consumption

Nowadays, data science and machine learning areas play a crucial role by performing predictions of energy with reliable historic of consumptions and other sensors data. This is crucial for decision makers to formulate an optimized plan to buy and sell energy ahead of time. According to many data scientists the decision criteria to select sensors to be added to the historic of consumptions consists on the sensors evidencing a higher correlation with the consumption variable. However, experts in the explainable artificial intelligence field study the positive and negative impact of the sensors to the prediction accuracy.. This paper analyzes the correlation of the sensors of a building to the consumption variable. Furthermore, explanations regarding the contribution of the consumption and sensors features to the forecasting performance are supported by an explainable artifial intelligence method named Local Interpretable Model Agnostic Explanation. The forecasted consumptions use two methods: Artificial Neural Network and K-nearest Neighbors.