Time-series classification using neural Bag-of-Features

Abstract

Abstract-Classification of time-series data is a challenging problem with many real-world applications, ranging from identifying medical conditions from electroencephalography (EEG) measurements to forecasting the stock market. The well known Bag-of-Features (BoF) model was recently adapted towards timeseries representation. In this work, a neural generalization of the BoF model, composed of an RBF layer and an accumulation layer, is proposed as a neural layer that receives the features extracted from a time-series and gradually builds its representation. The proposed method can be combined with any other layer or classifier, such as fully connected layers or feature transformation layers, to form deep neural networks for time-series classification. The resulting networks are end-to-end differentiable and they can be trained using regular back-propagation. It is demonstrated, using two time-series datasets, including a large-scale financial dataset, that the proposed approach can significantly increase the classification metrics over other baseline and state-of-the-art techniques.

Temporal Attention-Augmented Bilinear Network for Financial Time-Series Data Analysis

“…A more elaborate pipeline consisting of multiresolution wavelet transform to filter the noisy input series, stacked Auto-Encoder to extract high-level representation of each stock index and an LSTM network to predict future prices was recently proposed in [54]. Along with popular deep networks, such as CNN, LSTM being applied to time-series forecasting problems, a recently proposed Neural Bag of Feature (NBoF) model was also applied to the problem of stock price movement prediction [55]. The architecture consists of an NBoF layer which compiles histogram representation of the input time-series and an MLP that classifies the extracted histograms.…”

Forecasting jump arrivals in stock prices: new attention-based network architecture using limit order book data

“…This enables market analysis on a completely new level on many interesting questions (see, for example Toth et al, 2015;Chiarella et al, 2015), but has also brought unique challenges for both theory and computational methods (Cont, 2011). In the recent literature, both tractable models and data-driven approach-that is, machine learning-have been introduced to predict price movements with LOB data (Cont et al, 2010;Cont, 2011;Cont and De Larrard, 2012;Kercheval and Zhang, 2015;Ntakaris et al, 2018;Tsantekidis et al, 2017b,a;Passalis et al, 2017;Dixon, 2018;Tran et al, 2018;Sirignano and Cont, 2018). Overall, the existing literature provides evidence that limit order book data can be used to predict price movements in stock markets.…”

Multilinear class-specific discriminant analysis

“…While the performance of MCSDA was not better than its vector counterpart in the above face verification experiments, MCSDA outperforms all competing methods in the stock prediction problem, including the more complex neural network-based bag-of-words model N-BoF ( [27]). …”

“…Since FI-2010 is an unbalanced dataset with the mid-price remaining stationary most of the time, we cross-validated based on average f 1 score per class and also report the corresponding accuracy, average precision per class, average recall per class. Since our experimental protocol is the same with that used in ( [27]) for the Bag-of-Words (BoF) and Neural Bag-of-Words (N-BoF) models, we directly report their results. In addition, we report the baseline results from the database ( [26]) using Single Layer Feed-forward Network (SLFN) and Ridge Regression (RR).…”