Multivariate Time Series Classification Using LSTM
How to Run: main.py
and Multi-variate-Time-series-Data.xlsx
need to be on the same folder. The script will create four images
dataset Stats:
Multi-variate-Time-series-Data.xlsx
- Total Number of Time Series : 205
- Data For Training : 72%
- Data For Validation : 8%
- Data For Testing : 20%
Method Overview : I have used Keras framework and an LSTM Network to design the model
Train-Test Data Generation
win_size = 89
represents time series size. num_var = 14
represents the feature vector size. Based no this I extract segments
of 89*14
array. FUrtherr, we feed this time series to LSTM sequntially.
Parameter setting
split_ratio = 0.8
learning rate=0.001
nb_epoch=50
batch_size=64
Output:
- Performance of model on
test set
,
Test ROC | Test Confusion matrix |
---|---|
- Performance of model on
train set
,
Train ROC | Train Confusion matrix |
---|---|