Table 4 Most predictive features in EU and SA according to the Linear SVM with L1 penalty
From: Improving official statistics in emerging markets using machine learning and mobile phone data
In EU | |
1. | response_delay_text__weekday__allday__callandtext__max__mean |
2. | entropy_of_contacts__weekday__day__text__mean |
3. | percent_initiated_conversations__weekday__night__callandtext__mean |
4. | interevent_time__weekday__allday__text__max__std |
5. | percent_nocturnal__weekday__allday__text__std |
In SA | |
1. | percent_initiated_interactions__weekend__night__call__mean |
2. | percent_at_home__weekday__day__mean |
3. | call_duration__allweek__allday__call__max__mean |
4. | entropy_of_antennas__weekday__allday__mean |
5. | percent_initiated_interactions__allweek__allday__call__mean |