Artificial Bee Colony-Optimized LSTM for Bitcoin Price Prediction


Abba Suganda Girsang

Computer Science Department, BINUS Graduate Program – Master of Computer Science, Bina Nusantara University, Jakarta, Indonesia 11480
Background

  • Training LSTM model need optimal hyperparameters to obtain optimal model
  • Training LSTM model require long training time

    • Objectives

  • Design hyperparameter optimization on LSTM model using Artificial Bee Colony (ABC) Algorithm for Bitcoin Price Prediction
  • Implement hyperparameter optimization on LSTM model using Artificial Bee Colony (ABC) Algorithm for Bitcoin Price Prediction

    • Hyperparameter and ABC Parameter Settings
    # Hyperparameter Range Value Interval
    1 Sliding Window Size 40-70 5
    2 Number of LSTM neurons 30-100 5
    3 Dropout 0.3-0.5 0.01
    4 Learning Rate 0.0001-0.01 0.0001
    5 Regularizer L1, L2, L1L2 -
    6 Regularizer Rate 0.005-0.02 0.001
    7 Optimizer RMSProp, Adam, Nadam -
    # ABC Parameter Value
    1 Dimension 7
    2 Solution Number 10
    3 Population Size 20
    4 Limit 7
    5 Maximum Cycle Nuber 10
    Method

    Figure 1. Process of optimization
    Result
    # Best Hyperparameter Value
    1 Sliding Window Size 60
    2 Number of LSTM neurons 65
    3 Dropout 0.31
    4 Learning Rate 0.0091
    5 Regularizer L2
    6 Regularizer Rate 0.014
    7 Optimizer RMSProp
    # ABC-LSTM Measure Value
    1 Mean 206.8833373
    2 Standard Deviation 13.31792151
    3 Best 183.3421175
    4 Worst 240.8371386/td>

    Figure 2. ABC-LSTM

    Figure 3. LSTM
    Conclusions

    This research proved that ABC can be used as a method for optimizing hyperparameter for models that use LSTM for bitcoin price prediction. Optimized hyperparameter in this research consisted of sliding window size, number of LSTM units, dropout rate, regularizer , regularizer rate, optimizer and learning rate. Prediction result of ABC-LSTM outperformed LSTM prediction result.