Overview

Traffic jams in the city, as a result, many people in the city are choosing public transportation instead of their private cars. Therefore, scooters are used in response to their origin(first mile) and destination(last mile). This project aims to bring scooter user data to benefit the business.

Team

• gsuy
• AoFNontawat

Problem formation

• predict the next 24-hour of scooter pick-ups.
• predict trip destinations

Predict the next 24-hour of scooter pick-ups.

• preprocess with z-score normalization
• divide the data into 4 groups using jenks natural breaks

Model Structure

• model 1

• model 2

• model 3

Evaluate model

evaluate predictive models with time-based sliding window

• Overestimate = ( (actual of non-zero pick-up) - predict ) >= 0

• Underestimate = ( (actual of non-zero pick-up) - predict ) <= 0

• Zero accuracy = ( (predict zero pick-up) / (actual of zero pick-up) )*100

  
  
  
  

Predict trip destinations

• preprocess with min-max normalization
• API service

Model Structure

Evaluate model

evaluate predictive models with 10-fold cross-validation

• Average Distance Days of the week

• Average Distance hours of the Day