preprocessing

With the rapid spread of technology, large volumes of unlabeled data with large dimensions needed to be processed. To reduce the dimensions, unsupervised feature selection is known as an important pre-step before machine learning tasks. In this paper, an unsupervised feature selection method is proposed. The method works dynamically and is scalable based on matrix graphs and weighted matrices. To improve the performance of this method, instead of using the Lagrange function to construct a weight matrix, a bipartite graph theory is applied. Feature selection is done on the matrix graph. This graph is constructed using k nearest neighbors, which makes the method more robust to noise. The global structure of the original data is also preserved by constructing a Reconstruction Weight Matrix with low-rank constraint. In addition, the feature score, which explicitly reflects the strength of the features, is modeled using the Frobenius norm function. The proposed method is compared with similar methods in three criteria of classification accuracy, parameter sensitivity and complexity. Experiments show that the classification accuracy of the method presented in this paper has improved by an average of 2.83%. Its complexity has also been reduced to max{O(n2d),O(nm)}, where n is the number of samples, d is the number of features and m is the number of anchor points.

Electrical load forecasting is the prediction of future demands based on various data and factors containing different consumptions on weekdays, electricity prices and weather conditions that are different for societies and places. Generally, medium-term electrical load forecasting is often used for the operation of thermal and hydropower plants, optimal time planning for maintenance of power plants and the power grids. However, long-term electrical load forecasting is used to manage on-time future demands and generation, transmission and distribution expansion planning. In this paper, a hybrid long-term load forecasting approach using wavelet transform and an outlier robust extreme learning machine is proposed. Hourly load and temperature data were extracted from the GEFCOM 2014 database and divided into two classes of training and test. The one-level wavelet transform is used to decompose data to extract properties and reduce the dimensions of the data matrix. Decomposed low-frequency component (approximations) and high-frequency component values (details) from wavelet analysis are entered into the model for training and forecasting. For comparison accuracy of the proposed method, wavelet transform is applied to the data for the other three extreme learning machines. Also data without wavelet transform entered into four other forecasting models and the load forecasting results are compared with the proposed method. The results of the above mentioned evaluation show that electrical load forecasting by using wavelet transform and outlier robust extreme learning machine improves forecasting accuracy and the MAPE reduces to 3.0966. The overall calculated error by the proposed method was the best result obtained between the three several models of extreme learning machines and without preprocessing model. The MAPE is 0.4208 less than the ELM, 0.944 less than the RELM, and 0.1353 less than the WRELM model, respectively.

Recently, using social networks increases and people propagate their information through this networks. One of the most important challenges in these networks is sentiment attack, in which the attacker spreads rumors to influence users. Therefor rumor detection become important and attracts expanding research attention. Most of the previous works using deep neural networks for rumor detection without special preprocessing but we propose a new method for preprocessing data before learning which improve results. we use LSTM-CNN architecture with cyclical learning rate to detect Persian rumors. Beside that we investigate BERT model for Persian tweets. Our results demonstrate the effectiveness of this approach for English and Persian rumor detection.