<![CDATA[The goal of this study was to determine the optimal combination for optimizing the Patterned Dataset Model, particularly in patterned datasets during periods of price decline (crash). In previous research, the Crash Patterned Dataset has been shown to predict the next Bitcoin price. In this study, an experiment was conducted using a combination of prediction models, including ARIMA, machine learning, and deep learning. This research was conducted in 3 stages. The first stage is to compare the error results from the Bitcoin pair IDR crypto asset prediction process, which are part of the stored data from the patterned dataset under crash conditions. This dataset was tested with several prediction models, and the LSTM model with 60 seconds of resampling produced the best results, with an MAPE of 0.19%. In the second stage, BTCIDR, as part of the data from the patterned dataset in crash conditions, was resampled with variants 1D, 2D, 3D, 4D, 5D, 6D, 7D, 1H, 2H, 3H, 4H, 5H, 6H, 7H, 8H, 9H, 10H, 11H, and 12H. The result is that BTCIDR with a 3H resample has the lowest MAPE, at 1.3%. In the third stage, the prediction process is carried out using the LSTM model on the BTC IDR test dataset (as part of the Patterned Dataset in crash conditions) with a 3H resample. The dataset range is from May 2022 to 2025-01-23 11:05:48. This test predicts the Bitcoin IDR price series for the next 30 days, calculates the MAPE between the predicted series and the actual BTC IDR dataset 30 days later, and evaluates the results. The MAPE value for the Bitcoin IDR price prediction was 9.27%. This indicates that the average prediction error against the actual price is around 9.27%. The main objective of this research is to more accurately predict the price of the Bitcoin-IDR pair, providing additional helpful information for trading cryptocurrencies.]]>
Copyrights © 2025
