We implement a machine learning process to identify accounts that are, by our calculations, unlikely to engage.
‘Unlikely to engage’ percentage does not indicate suspicious activity on the part of the influencer. It is a calculation based on the usage patterns of the influencers audience in an effort to predict what percentage of their audience is unlikely to engage with their posts.
Our algorithm's classification is based on a decision tree, its error rate doesn't exceed 3.7% per test sample of 10,000 analysed accounts.
The main features used in the algorithm are the following:
- Followers/followings numbers and ratio
- Number of posts
- Account privacy
- Registration date
- Geotags usage
Our decision tree model and the selection of specific features are based on the Influencer Marketing industry expertise.