Neural Networks

Go Back to Home Page

A neural network is a highly effective algorithm for classifiying a labeled variable. The network predicts the outcome and then learns from its error--constantly improving its prediction.
I trained a neural network in Python. A great site to play with the theory behind neural networks is Neural Network Playground

Predicting Bike Rentals Using a London Bike Sharing Dataset

I used a dataset I found on Kaggle for this project Bike Sharing Dataset
I imported the file into a Jupyter notebook and started changing the columns. As we see below, I classified values above the median as '1' and below the median as '0'
Next, we want to train the neural network. After assigning the label field as the new count variable, we initialize the neural network and begin to training the network to predict the count variable.

The neural network can predict the label of the count variable with ~73% accuracy.

Impact

Since I fed in several key variables about time of day, season, and weather it is possible to adjust those variables when placing bikes in London to maximize bike sharing. There is a ton of potential in this project to help increase the availability of bike rentals and understand when and why bikes are rented.

Full code can be found here: https://github.com/basillatif/NeuralNetfoLondonBikeDataset