Inspiration
Time series forecasting models and be applied to numerous real-world problems across fields such as investment banking, Earth and planetary science, and biomedicine. With car sales data readily available, we took it as an opportunity to learn and practice building such models. We chose LSTM (an architecture based on RNN). LSTMs are very good at predicting sequential events as they is capable of capturing the patterns of both long term seasonalities such as a yearly pattern and short term seasonalities such as weekly patterns. Car sales numbers are known to have seasonal fluctuations, making LSTM an ideal choice of model.
What it does
Our LSTM model takes in cleaned data of car sales in the past from Stats Canada and predicts the number and value of cars will be sold in the future within a span specified by user.
How we built it
Data pre-processing, training, and tuning were done on Google Colab using Pandas, NumPy, Keras, Tensorflow as well as other supporting modules. Front-end development was done in Flask.
Future of our project
Models other than vanilla LSTM (e.g. simple RNN) could be implemented.
Log in or sign up for Devpost to join the conversation.