Probabilistic Forecasting of Multiple Time Series with Single Recurrent Neural Network

Time series forecasting can be summarized as predicting the future values of a sequence indexed by timestamps based on the past records of that sequence. Optimal or near-optimal resource allocation requires accurate predictions into the future. The study presents investigation performed on both classical methods and more contemporary methods from the literature. The classical methods studied are Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing (ETS) and Seasonal-Trend decomposition using LOESS (STL). One of the more contemporary time series models is a deep learning method called DeepAR, which is a Recurrent Neural Network (RNN) comprised of Long Short Term Memory (LSTM) cells. Both novel and classical approaches pose challenges unique to their own methodologies. Classical methods for example require the data to satisfy restricting modeling assumptions and each series to be preprocessed and modeled one by one. On the other hand, machine learning methods require a great amount of quality data for training, where it would be a challenge where the abundance of data may not be available. The proposed solution is to model many similar series with a single common model at the same time. The similarity in the context of this research refers to the fact that each series has the same recorded metric. The data set investigated in the study holds records of demand for many physical stores throughout Türkiye, where each individual series corresponds to a single physical location. There are a total of 120 individual series spanning between 2018 and 2022. The probabilistic forecasts are obtained by training the DeepAR model in order to learn a probability distribution and producing the point forecasts from sampling the learned probabilistic function. The probabilistic forecasts of different quantiles provide practicality such as forecasting in different quantiles for each series individually and can be tuned for different sensitivity to forecasting errors per series. At the end of the study, the provided error metrics for accuracy measurement are MAE (Mean Absolute Error) in order to show the actual value of the forecasted demand and MAPE (Mean Absolute Percentage Error) in order to compare the results with other models independent of the scale. The DeepAR model provided more accurate results compared to ETS and ARIMA on average for the whole data set in terms of both accuracy metrics.


Almost Periodic Solutions of Recurrent Neural Networks with State-Dependent and Structured Impulses
Akhmet, Marat; Erim, Gülbahar (2023-01-01)
The subject of the present paper is recurrent neural networks with variable impulsive moments. The impact activation functions are specified such that the structure for the jump equations are in full accordance with that one for the differential equation. The system studied in this paper covers the works done before, not only because the impacts have recurrent form, but also impulses are not state-dependent. The conditions for existence and uniqueness of asymptotically stable discontinuous almost periodic s...
A neuro-fuzzy MAR algorithm for temporal rule-based systems
Sisman, NA; Alpaslan, Ferda Nur; Akman, V (1999-08-04)
This paper introduces a new neuro-fuzzy model for constructing a knowledge base of temporal fuzzy rules obtained by the Multivariate Autoregressive (MAR) algorithm. The model described contains two main parts, one for fuzzy-rule extraction and one for the storage of extracted rules. The fuzzy rules are obtained from time series data using the MAR algorithm. Time-series analysis basically deals with tabular data. It interprets the data obtained for making inferences about future behavior of the variables. Fu...
A temporal neuro-fuzzy approach for time-series analysis
Yılmaz (Şişman), Nuran Arzu; Alpaslan, Ferda Nur; Department of Computer Engineering (2003)
The subject of this thesis is to develop a temporal neuro-fuzzy system for fore- casting the future behavior of a multivariate time series data. The system has two components combined by means of a system interface. First, a rule extraction method is designed which is named Fuzzy MAR (Multivari- ate Auto-regression). The method produces the temporal relationships between each of the variables and past values of all variables in the multivariate time series system in the form of fuzzy rules. These rules may ...
Global exponential stability of neural networks with non-smooth and impact activations
Akhmet, Marat (2012-10-01)
In this paper, we consider a model of impulsive recurrent neural networks with piecewise constant argument. The dynamics are presented by differential equations with discontinuities such as impulses at fixed moments and piecewise constant argument of generalized type. Sufficient conditions ensuring the existence, uniqueness and global exponential stability of the equilibrium point are obtained. By employing Green's function we derive new result of existence of the periodic solution. The global exponential s...
A Computational approach to detect inhomogeneities in time series data
Yazıcı, Ceyda; Yozgatlıgil, Ceylan; Batmaz, İnci; Department of Statistics (2017)
Detection of possible inhomogeneity within a series is an important problem in time series data. There are many sources from which inhomogeneity can be originated such as mean shift, variance and trend change, gradual change, or sudden decrease or increase in time series. Since time series has many application areas, the detection of changepoints should be investigated before conducting any analysis. Available methods have certain drawbacks that may lead to unreliable inferences. These include the need of i...
Citation Formats
S. T. TOPALLAR, “Probabilistic Forecasting of Multiple Time Series with Single Recurrent Neural Network,” M.S. - Master of Science, Middle East Technical University, 2022.