The whitepaper v1.2 has been released!
We are excited to provision the Big data, machine learning and AI are used as our technical stack for trend forecast.
When it comes to time-series forecasts, people always only use historical data on the variable to do forecasting. The traditional time-series forecast models like Autoregressive Integrated Moving Average (ARIMA) are more appropriate for univariate and stationary time-series data. But BTC prices are highly volatile, non-linear and non-stationary. Due to its rapidly changing nature, we need to add new features combine traditional machine learning models to forecast BTC prices. Our objective is to estimate the value of a target variable x in a future time point π₯Μ [π‘+π ] = π(π₯[π‘],π₯[π‘β1],β¦,π₯[π‘βπ]),π >0, s is the horizon for forecast. The first step, we would focus on short time forecast at the beginning, which means we take into consideration daily closing price forecast, and price increase/decrease forecasting for the short term (end-of-day and next day) as the horizon for forecast. Our long-term goal is to forecast 7β30 days. As for machine learning algorithms, we use the following ML models for classification and regression:
- support vector machines (SVM)
- artificial neural network (ANN)
- stacked artificial neural network (SANN)
- long short-term memory (LSTM)
The classification is applied as follows: If the BTC daily closing price ππ΅ππΆ[π‘+1]βππ΅ππΆ[π‘]β₯0 then π¦[π‘]=+1, and if ππ΅ππΆ[π‘+1]βππ΅ππΆ[π‘]<0, then π¦[π‘]=0, where y[t] is a target variable for categories of increasing and decreasing price. where y[t] is a target variable for categories of increasing and decreasing price. The regression models are used to predict BTC prices in a horizon of forecast for end-of-day and next day and will expend to 7β30 days for a long-term plan. For this part, we could get data from Coinmarketcap, Blockchain Info, etc.
Check more details at: https://tokentrend.github.io/whitepaper/AI.html