citybikes-algorithm

These algorithms, given the dataset of the user's unlocks,
its location, the time and the day of the week, try to predict
the station where the user will probably go.

Course: Recognition theories and techniques
School: University of Trento
Year: 2° (actually, anticipated third year course) .

The features they use are:

• user's location
• time
• day of the week

Datasets:

• one small (300 unlocks)
• one big (1134 unlocks)

Datasets' informations:

• departure stations' names
• arrival stations' names
• release time
• release date
• duration of the trip

[ note: the dataset provided by the company that owns
the service has only unlocks, not trips, that have to
be computated, matching differents unlocks. Moreover,
the unlocks history is incomplete. ]

Other information (calculated):

• location of the stations
  • calculated via API
• day of the week
• arrival time

Algorithms tried:

• Bayes
• KNN

Others:

• polyfit
  • try to use bayes on a continuous scale, instead of discrete