Paper Example on Time Series Analysis: Dependence Among Observations

Introduction

Standard machine data sets involve the extensive use of observations that involve time as part of the observation hence the name time series. The time series is common as it adds specific time-dependent dimensions in between observations. This time dimension assists in providing additional information about the systems, which is an explicit order hence, some form of dependence among the different observations. This structure of the introduction of different time series of observations is critical as it assists in the extensive description and forecasting of systems. Time series analysis involves understanding a data set that has a set of various time-dependent observations to create simple mathematical relationships and models that will provide modest relationships between the sampled data. On the other hand, time series forecasting involves making various predictions rise or fall of the statistical handling of the time series data. Data made from previous observations are used to predict the future performance of the system when subjected to various instances in time (McNally et al. 2018).

Is your time best spent reading someone else’s essay? Get a 100% original essay FROM A CERTIFIED WRITER!

Save 25%

Generally, this paper is mainly about predicting bitcoin value as well as predicting the stock market flow. Over the past years, a number of algorithms have been created for the purpose of stock market value prediction, but only a few algorithms have centered on predicting bitcoin price. To predict the bitcoin prices using time series, the paper will incorporate various information including existing neural systems RNN and LSTM and also study past bitcoin price data by gaining knowledge about its trends in both form of general and seasonal trends, RNN, forecasting models and existing methods of forecasting. The predicted bitcoin prices would assist greatly in understanding the risk factor associated with bitcoin investment and also the loosing and profiting probability from the investment (Azoff, 1994).

Introduction to object recognition and time series data sets

To critically understand the concept of object recognition and the various tools and algorithms used in this field such as conventional neural networks (CNN), gated recurrent unit (GRU), and long short-term memory (LSTM) then it is essential to discuss object detection comprehensively. Object-oriented programming is a programming paradigm which concentrates on the representation of various instances of code in the form of objects and classes for ease of retrieval or identification. An object is the simplest instance of a class which could be a different combination of variables, function, or even data structures comprising of strings and arrays of data that is referenced by an identifier. A class, on the other hand, involves a combination of objects in an extensible code-template that provides different levels of access to the objects as either protected, public, or private. Object detection, therefore, is a technology that is closely related to the object-oriented paradigm where the algorithms are specific in locating various instances of semantic objects of certain classes that could, for example, library books, buildings, marks pertaining students among others. These objects enable the building of different forms of artificial neural networks since they are based on the human nervous system.

The neural network is based on a collection and combination of various units of data called nodes which are built just like objects. The neural network is programmed to detect different kinds of objects from simple examples which are initially fed into their data systems. For example, the network will identify different types of books depending on their information, which is maybe the publisher name, number of pages, among others. This ability to identify and group objects depending on initialized characteristics is termed as data detection.

Conventional Neural Network (CNN)

This is a type of artificial neural network which has been a program to examine different kinds of visual representations specifically. The architecture includes an input and output layer as well as other multiple hidden layers which facilitate the conversion of a single volume activations through a differentiable function (Yin et al. 2017). The three major layers that are concerned with this specific network architecture include the convolution layer, pooling layer, and the fully connected layer. The layers maintain the different dimensional characteristics, which is the length, width, and the height hence a three-dimensional input. The convolution layer is responsible for the computational work and extracts all the features from the input image through a mathematical operation.

In summary, CNN will use its input layer to take in three-dimensional characteristics and feed it to the convolution layer which will calculate the output neurons that are involved with the fed input in terms of their local regions. The other layers, such as the pooling layer will downsample the input along the spatial dimensions in order to come up with the volume. In this way, this neural network converts the original image layer by layer from the original pixel to the final class scores.

Long short-term Memory (LSTM)

This is a special type of RNN which can learn and memorize long term dependencies between data. This is due to its ability to loop back and acquire the information that has already been processed, although it has already been worked on. This ability to loop back and get information makes it easier to forecast the information and time series due to the established dependencies among the data. In case there is a gap between the data, or in case there are some inconsistencies in the data, the LSTMs can identify them and develop a method to predict the consistencies as well as the irregularities. The architecture of the LSTMs involves the use and extensive application of the information about logic gates. The information to be manipulated is received from the input layer, which is then acted upon by the various gates and activation cells.

Gated Recurrent Unit

GRU, on the other hand, is a gating mechanism that has been recently developed that has better performance with smaller data sets compared to LSTM. The operating mechanisms of GRU are similar to that of LSTM but are more precise and only possess two gates compared to LSTM, which has four to six gates. GRU also lacks an output gate but only possess the input and update gates hence transfers information in the hidden state. The update gate, in this case, substitutes the input and forget gate that was present in LSTM hence the decision of which information to keep and which to let go. The reset gate, in this case, serves the purpose to decide the amount of the previous information to let go.

Predicting Bitcoin Price using RNN

Basing on the graph below depicting the historical data of bitcoin prices in the past years, the aspect of volatility can be noticed clearly. There exist various theories that relate to the exact reasons concerning the bitcoin prices volatility, and they are utilized in supporting the reasoning in crypto prices prediction in this case Bitcoin (Bouoiyour et al. 2016). There exist various approaches of predicting the cryptocurrencies future although, in this paper based on the brief explanation about multiple examples of RNN (recurrent neural networks) above, the main focus is on algorithmic trading (Zhang, 2003). That is using the past numerical data of bitcoin prices to train RNN to predict the future prices of bitcoin. Examples of RNN systems include CNN, LSTM, and GRU. According to the above definition about the three systems, LSTM is the better one on predicting time series data. This is because LSTM has the ability to learn and memorize long term dependencies between data due to its capability to loop back and acquire the information that has already been processed although it has already been worked on. Thus, its ability to loop back and get information makes it easier to forecast the information and time series due to the established dependencies among the data.

RNN can be defined as a classification of artificial neural networks at which nodes connections form a graph that is directed along a sequence (Husken & Stagge, 2003). The RNN networks depict temporarily dynamic behavior of a sequence basing on time and can utilize its state internally to process sequences. Practically, one can achieve this sequential process using the layers of GRU and LSTM (Fu et al. 2016). In this case, due to the use of a time series dataset, the feedforward-only neural network is not practicable if used as the prices of bitcoin tomorrow is correlated mostly a month's ago and today's prices thus the use of RNN, in particularly, LSTM.

The steps to be followed to enable the creation of a specific program that trains on the past data prices of bitcoin and facilitates the prediction of the future price of bitcoin is outlined below. By following the steps below in depth and a length, a developed and trained RNN sequential model is obtained and can successfully predict the future prices of bitcoin and even saved to be used for more prediction purposes for example prediction of the stock market (Georgoula, 2015). The trained model can be utilized in either mobile or web application by redirecting to OOP (Object Oriented Programming). The created program is generally essential to the blockchain, artificial intelligence, and finance.

• Obtain, Clean, and normalize the Bitcoin price historical data.
• Build RNN (Recurrent neural network) with LSTM (Long short-term memory) (Jang & Lee, 2017)
• RNN training and save the trained sequential model
• Predict future bitcoin prices using the saved trained model and deserialize it.

Conclusion

In conclusion, predicting bitcoin future prices using LSTM is effective since, among the other neural systems-CNN and GRU, LSTM is better in predicting time series data thus better results are expected (Sergios Karagiannakos, 2018). One disadvantageous factor is the use of past data as the best approach of predicting the future bitcoin price as they may be associated with absolute misunderstanding however with the help of other architectures, for example, convolutional networks; there is a potential.

References

Azoff, E. M., (1994). Neural network time series forecasting of financial markets. John Wiley & Sons, Inc.

Bouoiyour, J., Selmi, R., Tiwari, A. K., & Olayeni, O. R. (2016). What drives Bitcoin price. Economics Bulletin, 36(2), 843-850.

Fu, R., Zhang, Z., & Li, L. (2016, November). Using LSTM and GRU neural network methods for traffic flow prediction. In 2016 31st Youth Academic Annual Conference of Chinese Association of Automation (YAC) (pp. 324-328). IEEE.

Gazi., O., (2018). Predict Tomorrow's Bitcoin (BTC) Price with Recurrent Neural Networks. Retrieved from; https://towardsdatascience.com/using-recurrent-neural-networks-to-predict-bitcoin-btc-prices-c4ff70f9f3e4

Georgoula, I., Pournarakis, D., Bilanakos, C., Sotiropoulos, D., & Giaglis, G. M. (2015). Using time-series and sentiment analysis to detect the determinants of bitcoin prices. Available at SSRN 2607167.

Husken, M., & Stagge, P., (2003). Recurrent neural networks for time series classification. Neurocomputing, 50, 223-235.

Jang, H., & Lee, J. (2017). An empirical study on modeling and prediction of bitcoin prices with Bayesian neural networks based on blockchain information. IEEE Access, 6, 5427-5437.

McNally, S., Roche, J., & Caton, S. (2018, March). Predicting the price of Bitcoin using Machine Learning. In 2018 26th Euromicro International Conference on Parallel, D...