BitForecast is a simple project based on Time Series Data Analysis which helps to predicts/forecasts the value of bitcoin in the near future. It consists of several models which calculates the value of bitcoin. The dataset is taken the CoinDesk.
BitForecast have different models implemented which are as follows:
| EXPERIMENTS | MODEL |
|---|---|
| Model 0 | Naive Bayes Model (baseline) |
| Model 1 | Dense Model (Window size=7, Horizon = 1) |
| Model 2 | Dense Model (Window size=30, Horizon = 1) |
| Model 3 | Dense Model (Window size=30, Horizon = 7) |
| Model 4 | Conv1D Model |
| Model 5 | LSTM Model |
| Model 6 | Dense Model (multivariate data) |
| Model 7 | N-BEATS Algorithm |
| Model 8 | Ensemble Model (multiple models stacked) |
| Model 9 | Future Prediction Model |
| Model 10 | Dense Model (turkey data) |
- N-BEATS algorithm
- Ensemble model
- Future Prediction model
- Model performing on Turkey data
- Comparing results of each model
-
Naive Bayes Model (Baseline):
-
Future Prediction Model
-
Prediction on Turkey Dataset:
-
Comparing Results:
The Baseline model is able to perform the best among all other models achieving the highest accuracy.