I am trying to classify a series of timestamps using RNN with LSTM. The data consists of timing information extracted from the uplink packets recorded during a website fetch. The dataset contains 100 individual fetch samples of each website in a set of 100 websites, this gives me 10,000 samples.

I would like to teach a neural network to classify sequences of timestamps to which website they belong.

An example sequence would look like:

0.0, 0.25420099, 0.70250899, 0.7434534, 0.8746745, ... 2.54634634

Each of these values represents an offset from the start of the website fetch.

These all differ in lengths and are between ~300 and ~4500 timestamps.

I have tried training the LSTM network in Keras like below:

modelClass = Sequential()
modelClass.add(Dense(1, activation='sigmoid'))
modelClass.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
modelClass.fit(X_tr, y_tr, epochs=3, batch_size=64)
scores = modelClass.evaluate(X_tst, y_tst, verbose=0)
print("Accuracy: %.2f%%" % (scores[1]*100))

Training data is in the shape of (9000, maximum_sequence_length, 1), however this gave me very bad results. I am new to machine learning and don't fully understand how to find appropriate algorithms for specific tasks. Google searches did not come up with any ways to classify such data.

I would appreciate any suggestions and directions, please let me know if more information is required.


Your Answer

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

Browse other questions tagged or ask your own question.