ML4MQ : Machine Learning For Mobile QoS¶
Title: ML4MQ-2.0 - Mobile User Throughput Prediction
Authors: Philippe DOOZE, Alassane SAMBA
Date : 03/2017
ML4MQ (Machine Learning For Mobile QoS) is an Atom able to estimate throughput class reachable by a mobile user on a 4G network without any data consumption. The mobile throughput class are the following.
- Class 0 : Lower or equals to 15 000 Kbit/s
- Class 1 : Strictly higher than 15 000 Kbit/s and lower or equal to 30 000 Kbit/s
- Class 2 : Strictly higher than 30 000 Kbit/s and lower or equal to 45 000 Kbit/s
- Class 3 : Strictly higher than 45 000 Kbit/s
ML4MQ is based on a predicitve model that needs the following data to be able to estimate a throughput class for a specific user.
- 1 : User's ID
User’s radio context features:¶
- 2 : cell Band : Frequency band of the cell
- 3 : RSRP : Reference signal Received Power
- 4 : RSRQ : Reference Signal Received Quality
- 5 : SNR : Signal Noise Ratio
Radio Access Network performance features:¶
- 6 : avg_down_throughput : The user's downlink throughput average in the serving cell.
- 7 : nb_rrc_attempt : The number of RRC connection request in the serving cell.
- 8 : CSSR : Call Setup Succes Rate in the serving cell .
ML4MQ must be launch under spark environement thanks to the following command :
- spark-submit launch_ml4mq_module.py input_data model_dir output_dir
Here, the arguments are as follows:
- input_data-> data awaited by the module for throughput prediction
- model_dir -> directory of the provided model
- output_dir -> directory to store the results