1
$\begingroup$

Currently, I'm working on 6-axis IMU(Inertial Measurment Unit) dataset. This dataset contain 6 axis IMU data of 7 different drivers. The Imu sensor attached on vehicle. The drivers drives same path. So, the dataset include 6 feature columns and a label column.

I tried multiple neural network models.The sensor data is a sequential data so I tried LSTM(Long Short Term Memory) & classical fully-connected layers. Some of my architecture(in keras framework):


Layer (type)                 Output Shape              Param #   

lstm_4 (LSTM)                (None, 1, 128)            69120     
_________________________________________________________________
lstm_5 (LSTM)                (None, 1, 64)             49408     
_________________________________________________________________
lstm_6 (LSTM)                (None, 1, 32)             12416     
_________________________________________________________________
dense_8 (Dense)              (None, 1, 64)             2112      
_________________________________________________________________
dropout_2 (Dropout)          (None, 1, 64)             0         
_________________________________________________________________
dense_9 (Dense)              (None, 1, 7)              455       


2nd Architecture:

=================================================================
dense_10 (Dense)             (None, 32)                224       
_________________________________________________________________
dense_11 (Dense)             (None, 64)                2112      
_________________________________________________________________
dense_12 (Dense)             (None, 128)               8320      
_________________________________________________________________
dense_13 (Dense)             (None, 256)               33024     
_________________________________________________________________
dropout_3 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_14 (Dense)             (None, 512)               131584    
_________________________________________________________________
dense_15 (Dense)             (None, 256)               131328    
_________________________________________________________________
dense_16 (Dense)             (None, 128)               32896     
_________________________________________________________________
dense_17 (Dense)             (None, 64)                8256      
_________________________________________________________________
dropout_4 (Dropout)          (None, 64)                0         
_________________________________________________________________
dense_18 (Dense)             (None, 128)               8320      
_________________________________________________________________
dense_19 (Dense)             (None, 7)                 903     

The best accuracy in my models was %70 which is not good. How style of layers should I use to handle this data? Or, which type of model would increase accuracy?

$\endgroup$
2
$\begingroup$

after reading some literature in the area I'd recommend the following:

  • Try using Convolution Neural Networks (CNNs) this paper outlines some really good points on why should you use CNNs.
  • Try a Combination of the different layers in the same model. Start with some Convolution Layers, then some LSTMs and then a couple of Dense Layers followed by Dropout
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.