Feature selection is a widely studied technique whose goal is to reduce the dimensionality of the problem by removing irrelevant features. It has multiple benefits, such as improved efficacy, efficiency and interpretability of almost any type of machine learning model. Feature selection techniques may be divided into three main categories, depending on the process used to remove the features known as Filter, Wrapper and Embedded. Embedded methods are usually the preferred feature selection method that efficiently obtains a selection of the (...) most relevant features of the model. However, not all models support an embedded feature selection that forces the use of a different method, reducing the efficiency and reliability of the selection. Neural networks are an example of a model that does not support embedded feature selection. As neural networks have shown to provide remarkable results in multiple scenarios such as classification and regression, sometimes in an ensemble with a model that includes an embedded feature selection, we attempt to embed a feature selection process with a general-purpose methodology. In this work, we propose a novel general-purpose layer for neural networks that removes the influence of irrelevant features. The Feature-Aware Drop Layer is included at the top of the neural network and trained during the backpropagation process without any additional parameters. Our methodology is tested with 17 datasets for classification and regression tasks, including data from different fields such as Health, Economic and Environment, among others. The results show remarkable improvements compared to three different feature selection approaches, with reliable, efficient and effective results. (shrink)

This work presents a novel approach to forecast streaming big time series based on nearest similar patterns. This approach combines a clustering algorithm with a classifier and the nearest neighbours algorithm. It presents two separate stages: offline and online. The offline phase is for training and finding the best models for clustering, classification and the nearest neighbours algorithm. The online phase is to predict big time series in real time. In the offline phase, data are divided into clusters and a (...) forecasting model based on the nearest neighbours is trained for each cluster. In addition, a classifier is trained using the cluster assignments previously generated by the clustering algorithm. In the online phase, the classifier predicts the cluster label of an instance, and the proper nearest neighbours model according to the predicted cluster label is applied to obtain the final prediction using the similar patterns. The algorithm is able to be updated incrementally for online learning from data streams. Results are reported using electricity consumption with a granularity of |$10$| minutes for 4-hour-ahead forecasting and compared with well-known online benchmark learners, showing a remarkable improvement in prediction accuracy. (shrink)