Bike-Sharing-Prediction-Model

The project

Building a predictive model for the total number of hourly bike rentals in the bike sharing network of Washington, D.C. Python was the programming language for this project.

The code

The code is organized as follow: an exploration of the data, a k-fold cross-validation error to determine the best hyper-parameters of my model, and the predictions.

The data

The dataset provided was hourly rental data spanning two years with an exception of 2 months used as a testing set for predictions. I predicted the total count of bikes rented during each hour for the months of August and December 2012.

Part 1. Exploration of the data

Looked at the features of the data as well as the existence of missing values.

Part 2. Building of the Model

Built a Random Forest Regressor after having tuned it with a grid search cross-validation.

Part 3. Predictions and model performances

Evaluated the model performances and predicated the total count of bikes rented in the test set.

Results

Ranked first in a private Kaggle competition (22 participants).