A Hybrid Machine Learning Model for Clustering and Prediction of Closing Price of Cryptocurrency (Published)
The Present financial system operates on technological innovations with attention to cryptocurrencies and its related processes. Blockchain technology has increased exponentially since the release of Bitcoin in year 2008 as the first viable decentralized cryptocurrency. In recent years, research on machine learning, blockchain technology, and interaction has improved significantly. This work considers utilizing clustering and Machine Learning techniques to predict the closing price of cryptocurrencies. The study adopts K-means, Model Based Clustering Algorithms and Artificial Neural Network (ANN) approaches to implement clustering and time-series predictions. The capability of the Neural Network to provide one-day closing price prediction of Bitcoin was evaluated. The paper espouses a training ratio of 70:30 on the dataset deployed in the work for an increased accuracy due to large number of data-points. Basically, using K-Means and Model-Based Clustering algorithms indicated 74.5% accuracy based on the Elbowing method adopted for the determination of optimal number of clusters in the data set. Increasing the number of clusters to 10 in the data points demonstrates an accuracy of 88.4%. The Empirical findings reveal that, Mean Squared Error (MSE) score shows 1.20, the Mean Absolute Error (MAE) score illustrate 2.9 on the test dataset after evaluation of the prediction model. A comparative analysis shows the advantages of using K-Means, Model-Based Clustering algorithms and artificial neural network to provide trustworthy, automatic monitoring and clustering as well as prediction. This further reveals that it is feasible to produce an estimation for which price moving indicators can impact the actual coin closing price operations.
Keywords: Cryptocurrencies, artificial neural network (ANN), blockchain technology, k-means cluttering, machine learning (ml), model-based clustering algorithms