The fast-evolving the cryptocurrencies present both the special opportunities and the problems. The risk involved in investing in the cryptocurrency assets is very high because the prices of the exchange can change in a day-to-day basis. The effort uses powerful machine learning techniques to forecast the value of cryptocurrencies. In comparison to the two other seven models with less faults, the Neutral Networks had the best forecast and validation efficiency. In order to predict future tendencies, LSTM (long short-term memory) neural networks were used. Complicated association of financial data can be analyzed using LSTM model. Overall, more than fifty cryptocurrencies were submitted to the Exploratory Data Analysis (EDA), which began with the gathering of historical data and continued with feature engineering, integrative binning, and data preparation and standardization. The most successful ones were identified by the price movement, market size, and volumes. The LSTM based model is coded in Python and the model has been applied on the 90-day data of price movements to check the existence of complicated patterns and correlations. The performance indicators to monitor the model performance were RMSE and MAE. These results corroborate the Adaptive Market Hypothesis (AMH), which posits that alterations in investor and market behavior have an impact on the dynamic efficiency of cryptocurrency markets. As shown in the paper, machine learning models have the potential in financial economics and the models will be beneficial in investment decision-making process and approaches of risk management.

Keywords: Machine learning, financial economics, LSTM neural networks, model prediction, and currency forecasting