Anomaly Detection in Bitcoin Prices using DBSCAN Algorithm

Blockchain is an emerging technology which is also behind the Bitcoin digital money. Daily bitcoin transactions are increasing due to the popular and widespread investments. The increase of Bitcoin related datasets and this increased big dataset requires novel approaches and methods to analyze using data mining techniques. In addition, fluctuations and anomalies in the bitcoin prices could mean a great deal to economists and discovering anomalies in bitcoin prices is important. In this study, anomaly detection in Bitcoin prices is performed based on the change of Bitcoin price difference and the change of Bitcoin price difference in percentage with respect to previous day using 8-years of Bitcoin price dataset of the period of 2012-2019. First, the dataset is pre-processed and unnecessary columns are deleted. Then, 2 different datasets are created by using daily bitcoin prices, i.e. bitcoin price difference dataset and bitcoin price difference in percentage dataset. After that, for detecting anomalous price changes, DBSCAN algorithm and statistical method are used, and the performance of the algorithms are evaluated. The results show that the DBSCAN algorithm and statistical method successfully detects anomalies in bitcoin prices for both of the datasets. However, the DBSCAN algorithm performs better than the statistical method which could detect anomalies even they are close to the normal daily price changes. Also, in this study, bitcoin price difference dataset and bitcoin price difference in percentage dataset are compared and the differences of the results for both datasets and their reasons are explained.