Western Developer Society
DATE
October 2024 - Present
DESCRIPTION
Collected data on LTC Bus Transit for use in the model. Utilized data from LTC Real-Time GTFS Data, and weather data integration.
Utilized numerous data preprocessing processes such as Feature Engineering, Data Augmentation, Feature Scaling. Utilized temporal features of the data, day, day of year, hour, normalized all features in a range of [0, 1]. Utilized 20 time steps for the LSTM.
Utilized different training processes, 80/20 train-test split, batch size of 64, 200 training epochs with early stopping to prevent overfitting, model checkpointing to save best weights.
Trained and tested numerous different deep learning ML models, mainly LSTM's, to create the ML model that could give the most accurate predictions. Utilized an input layer, three Bi-LSTM layers with a dropout rate of 0.2 to prevent overfitting. The Bi-LSTM structure also allowed the model to process time sequences in both forward and backwards directions allowing the model to learn correlations in the data more effectively.
TECHNOLOGIES
PyTorch
Numpy
SciKit
Flask
