cluster-lab | Hypriot Cluster Lab allows to create a Docker | Continuous Deployment library

i am working on the classification(positive/negative) of the followers of a twitter account based on the followers tweets ,

collecting data

1. got all the followers and the their tweets from the respective account
2. sentiment analysis of each tweet and labelled as (positive/negative/neutral)

which one of these are right approach?if no... is there any better approach?

my approach 1 for user classification:

1. since polarity score of neutral is 0 ,delete all neutral labelled tweets
2. took the count of positive and negative tweets
   1. there are only 17% of negative tweets out of the all tweets(pos+neg)
3. i kept the threshold as 34% and grouped tweets based on the data user
   1. out the total tweets(pos+neg) by the user,if the negative tweets account more than 34% i am classifying him as the negative user otherwise positive user
4. Results:out of 300 followers i got 19 of them are negative rest of them are positive

my code for approach 1

I am using some default mapbox clustering, which is not enough for indoor map of building. Here is my example