Citibike-Ride-Duration-Neural-Network

Key points in this project:

• Purpose: Build a model that predicts bike ride duration with minimal error while operating within the limits of computing power.
• Framework: The modeling framework for this work is very simple and the target can be directly modeled.
• Randomness: The predictive variables in the dataset to be predicted and evaluated were chosed subjectively based on relevance to the outcome variable.

Key Strategies

• Restrict the training data. First restricted outcome variable of the tripduration to be from 1 minute to 1 hour, because less than 1 minute possibly means that the rider just tried on the bicycle and did not even take any ride, and more than 1 hour possibly means that the rider forgot to lock out the bicycle, which are all bad data to analyze.

Model

The model used in this work is a Linear Regression Model

Things I learned:

• Think the problem from a grand view, using different strategies to define the target and built up modeling framework.

Things could have done better.

• It would be good if I could try other smoothing techniques except for spline.