Training Database Description

To convert from the trace format to databases suitable for training and testing, we used a script that concatenates each day of the month. Then, based on the windows lengths, the script generates a database like the one partially shown in Table 9.2. While scanning the month, we report the maximum value of each traffic flow type. All the traffic data is then normalized by this value. The Time variable is simply a counter incremented at each time step. It represents the time of the day and it is normalized by the total number of samples per day. Similarly the Day represent the day of the week. It is normalized by 7. For the following experiments we used the total amount of traffic (the sum of the incoming and outgoing traffic). See Section 9.3.7 for details about the possible uses of the separate flows.

Table 9.2: Some entries from the training database after being normalized between 0 ...1.

Step	Day	Time	F(T)	F(T-1)	F$_y$(T)	F$_y$(T-1)	F$_w$(T)	F$_w$(T-1)	F(T+1)
...	...	...	...	...	...	...	...	...	...
21	0.000	0.037	0.406	0.500	0.024	0.029	0.002	0.001	0.442
22	0.000	0.038	0.442	0.406	0.029	0.024	0.002	0.002	0.372
23	0.000	0.040	0.372	0.442	0.028	0.029	0.003	0.002	0.229
24	0.000	0.042	0.229	0.372	0.031	0.028	0.009	0.003	0.166
25	0.000	0.043	0.166	0.229	0.032	0.031	0.009	0.009	0.212
...	...	...	...	...	...	...	...	...	...
743	0.143	0.240	0.104	0.173	0.298	0.311	0.002	0.002	0.087
744	0.143	0.242	0.087	0.104	0.280	0.298	0.003	0.002	0.093
745	0.143	0.243	0.093	0.087	0.314	0.280	0.003	0.003	0.281
746	0.143	0.245	0.281	0.093	0.256	0.314	0.003	0.003	0.411
747	0.143	0.247	0.411	0.281	0.238	0.256	0.003	0.003	0.049
...	...	...	...	...	...	...	...	...	...
1950	0.429	0.252	0.277	0.280	0.045	0.035	0.420	0.671	0.206
1951	0.429	0.253	0.206	0.277	0.053	0.045	0.165	0.420	0.221
1952	0.429	0.255	0.221	0.206	0.076	0.053	0.258	0.165	0.521
1953	0.429	0.257	0.521	0.221	0.134	0.076	0.360	0.258	0.101
1954	0.429	0.258	0.101	0.521	0.135	0.134	0.223	0.360	0.143
...	...	...	...	...	...	...	...	...	...