Order Now

Support Vector Machine for Cryptocurrency Forecasting

  • 28th Jul, 2020
  • 18:40 PM

SVM or Support Vector Machine is considered to be a convenient tool for data cataloging, which uses a non-leaner and non-parametric system. It uses a high dimensional place to search for a hyperplane, targeting maintain minimal error rate. The present paper focuses on the use of SVM to forecast the fiscal drive of cryptocurrencies, especially the Bitcoin. The growth of Bitcoin prices between 2016-2018 has experienced a dramatic hike. Therefore, forecasting cryptocurrencies became essential to increase the performance of the stock market (Akyildirim, Goncu & Sensoy, 2018). Accordingly, the use of machine learning and SVM becomes necessary to integrate to obtain a greater forecast. Therefore, the present paper will evaluate the issues and effects of SVM, and sentiment analysis used to forecast financial movement.

Research Objectives 

The research objectives are as follows:
1.    Explore existing research on SVM in financial forecasting.
2.    Gather sentiment-based dataset used for cryptocurrencies in social media.
3.    Develop a sentiment-based SVM that is optimized by Particle Swarm Optimization (PSO) for cryptocurrency forecasting algorithms.
4.    Evaluate the sentiment-based SVM for cryptocurrency forecast.
5.    Compare the performance of sentiment-based SVM optimized by PSO with benchmarked algorithms.


Research Questions 

Looking at the research objectives above, the following research questions have been made:
RQ1: Describe the drawbacks or limitations of the support vector machine used in financial forecasting.
RQ2: Demonstrate the effects of sentiment on social media towards financial forecasting.
RQ3: In which process sentiment can enhance the support vector machine?
RQ4: Does the performance of optimized PSO with sentiment improved the cryptocurrency forecasting compared to other machine learning algorithms


Literature Review 

Each type of classification system has some drawbacks and some advantages, which are significant conferring to the nature of data being scrutinized. Hence, it can be said that the advantages or drawbacks have comparative applicability. SVMs are suitable for liquidation analysis where the data is non-regulatory. For instance, data with an unidentified supply are efficiently assessed with SVM (Auria & Moro, 2008). However, the most significant drawback imposed by the SVM technique is the lack of limpidity in the results. The SVMs are incapable to signify the tallies of all companies as a generalized parametric function of the economic ratios, as the dimensions become very high. Loads of the economic ratios are not perpetual; hence, the peripheral input for each financial ratio to the tally becomes inconstant (Zaini, et.al 2018). Consequently, while using for financial anticipation, the size of the business or companies turn into a great disadvantage for SVMs. In order to process bigger datasets, the SVM system necessitates larger time to process, and it does not perform as expected in the cases of overlain classes. Another issue with SVM is the assortment of the suitable kernel is very intricate; thus time-consuming and increases the potentials of risks while making the estimates (Zaini, et.al 2018).

Researchers worldwide found different ways to make financial estimates. Hence, using sentiment analysis method, it is possible to make financial forecasting. However, it is perplexing to model a dynamic system to precisely calculate the variations in the stock market. Sentiment analysis is a method or an area of NLP or Neutral Language Processing, which looks to classify all the destructive and constructive feedbacks, responses, and approaches conveyed by social media users. Hence, the procedure of sentiment analysis hinges principally on different machine learning systems, for instance, Support Vector Machine to categorize the negative and positive classes. According to He, Guo, Shen & Akula, (2016), gathered data from social media can be considered as a sign of brand assets. The article examines the association between Twitter messages and fiscal stock prices. Hence, in the multivariate regression, the concurrent sentiments of different companies have been analyzed. The sentiments on Twitter has been found to be suggestively linked to the forthcoming stock prices of the selected companies. Negative sentiments on social media are unfavorable for the growth of stock prices. It has been recognized that a 1% upsurge in the negative response on Twitter resulted in a 3.7% fall in the stock price (He, Guo, Shen & Akula, 2016).

However, there are many prior researchers who claim that sentiments on social media do not affect the stock prices. But it is ostensible that sentiment analysis provides greater insight into the needs of customers as well as the market feasibility of a product, which is required by both researchers and investors (He, Guo, Shen & Akula, 2016). 

Sentiment analysis can be done in two ways, in which the MLSA (Machine Learning Sentiment Analysis) and the LBSA (Lexicon-based Sentiment Analysis). Hence, developing and enhancing the programmed techniques of sentiment analysis is significant (Jadav & Vaghela, 2016). A large amount of data is regularly gathered from different social networks and other data sources, which are useful to procure the fiscal estimates related to cryptocurrency. There are different methods, which can be utilized with SVM to enhance the performance of data classification. Feature extraction is a process of converting the input data set into a set of features. The performance of the MLSA therefore fundamentally depends on the features of data. Support Vector Machine typically works for text-based sentiment analysis by classifying the textual data gathered from different sources. Therefore, using pre-processing, training and post-processing strategies for sentiment analysis, the performance of SVMs can be improved. The strategies include optimization by incorporating data, which are interrelated to the number of classes present in the dataset (Zainuddin & Selamat, 2014).

The forecasting algorithms of Support Vector Machine can be used to organize the financial performance data related to cryptocurrencies, such as Bitcoin. The dataset of Bitcoin includes different features, such as market price, block size, cost per transaction, the revenue of the Bitcoin miners, market supply, output volume, trade volume, etc. Hence, the SVM algorithm is used as a machine learning model to forecast the price of Bitcoins (Raghava-Raju, 2018).

There are different machine learning processes that are used for the cryptocurrency forecasts, such as support vector machine, ANN or Artificial Neural Networks, and Deep Learning. SVM is an algorithm of machine learning that is considered to be the most flexible method to make clear and precise boundaries (Raghava-Raju, 2018). It functions very well in several applications used in the cryptocurrency forecasting and offers fast and reliable results compared to other MLs. However, the main drawback of SVM is, it is incapable of managing very large data in a less amount of time. The benefit of SVM is that it outperforms in the simplification model and performs very well with reduced datasets (Hitam & Ismail, 2018).

Another common technique is ANN, which comprises an input layer, a hidden layer, and output layer. It tends to work like a human brain to sense and analyze the data. However, it is a more complicated system, which is not cost-effective and requires a lot of time. Deep learning is considered to be a diverse method of machine learning. It is capable of producing greater results with the help of multiple layer extraction (Hitam & Ismail, 2018). The following table shows a performance evaluation of different technologies to predict different cryptocurrencies including Bitcoin.(Source: Hitam & Ismail, 2018)



The market of cryptocurrencies draws attention because of the recent progressions of different fundamental technologies. The technologies of Machine Learning such as Support Vector Machine, Neural Network, etc. helps to analyze the market and estimates the forthcoming changes in the stock market. Therefore, the investors can consider the techniques as a method of understanding the market and increase the consideration to the investment community. Evaluating the different factors of SVM, it can be concluded that it is the best and strong system amongst other techniques. From the overall results, it can be said that the variation rate of SVM low with a better proficiency of simplification, which helps to maintain a consistency of more than 50%. On different occasions of prediction, SVM can achieve more the 70% accuracy rate, which indicates that the advanced methods of machine learning are highly promising for short-term prediction of cryptocurrency market trends. 


  • Akyildirim, E., Goncu, A., & Sensoy, A. (2018). Prediction of Cryptocurrency Returns using Machine Learning, 2-24.
  • Auria, L., & Moro, R. A. (2008). Support vector machines (SVM) as a technique for solvency analysis, 1-16.
  • Han, S., & Chen, R. C. (2007). Using svm with financial statement analysis for prediction of stocks. Communications of the IIMA, 7(4), 8.
  • He, W., Guo, L., Shen, J., & Akula, V. (2016). Social media-based forecasting: A case study of tweets and stock prices in the financial services industry. Journal of Organizational and End User Computing (JOEUC), 28(2), 74-91.
  • Hitam, N., & Ismail, A. (2018). Comparative Performance of Machine Learning Algorithms for Cryptocurrency Forecasting. Indonesian Journal Of Electrical Engineering And Computer Science, 11(3), 1121. doi: 10.11591/ijeecs.v11.i3.pp1121-1128
  • Jadav, B. M., & Vaghela, V. B. (2016). Sentiment analysis using support vector machine based on feature selection and semantic analysis. International Journal of Computer Applications, 146(13).
  • Raghava-Raju, A. (2018). A Machine Learning Approach to Forecast Bitcoin Prices. International Journal of Computer Applications, 182(24).
  • Zaini, N., Malek, M. A., Yusoff, M., Mardi, N. H., & Norhisham, S. (2018, April). Daily River Flow Forecasting with Hybrid Support Vector Machine–Particle Swarm Optimization. In IOP Conference Series: Earth and Environmental Science (Vol. 140, No. 1, p. 012035). IOP Publishing.
  • Zainuddin, N., & Selamat, A. (2014, September). Sentiment analysis using support vector machine. In 2014 International Conference on Computer, Communications, and Control Technology (I4CT) (pp. 333-337). IEEE.

Share this post