fle1scha/Deep-Learning-Network-Traffic-Prediction

DL4NTP is an Honours-level research project investigating LSTM neural networks as network traffic predictors on the South African National Research Network ## Deep Learning for Network Traffic Prediction ### Using LSTMs as Network Traffic Predictors on SANReN #### Antony Fleischer and Justin Myerson The Deep Learning for Network Traffic Prediction (DL4NTP) system is split into two files: DL4NTP.py, which preprocceses the data and implements the models; and dataAnalysis.py, which extracts data and metrics from the models so that the results can be evaluated. 1. To run the DL4NTP.py and dataAnalysis.py program, a user should have the following packages installed: ``` import tensorflow as tf import math import matplotlib.pyplot as plt import numpy as np import os import pandas as pd import datetime as dt import time import csv import seaborn as sns import keras.backend as K ``` 2. To run either file, enter the following command in a terminal application. The user must have the SANREN.txt file in the same directory as the program. ``` python3 DL4NTP.py ``` or ``` python3 dataAnalysis.py ``` NOTE: the dataAnalysis file will not run succesfully if the DL4NTP program has not completed LSTM model training. 3. The user will be asked for prompts in the command line as the program run. Example answers to each input are provided below: ``` Would you like to view the preliminary statistical analysis plots? [Y/N] Y ``` When 'Y' is inputted, various graphs will pop up as the preliminary statical analysis methods execute. Please close each figure to continue. ``` View the split of training and test data? [Y/N] Y ``` ``` You will now be asked to define the ranges of the LSTM hyperparameter grid search. Please enter the first epoch gridsearch hyperparameter value: 5 Please enter another epoch gridsearch hyperparameter value or type DONE to continue: DONE ``` As the model training process is only run once, the range of hyperparameters that the grid search uses are defined at the beginning the program, whilst the optimal model selection hyperparameters are defined through user input after training. The DL4NTP study used a hyperparam …