Neural Network Library and Taxi Trip Duration Prediction

• Developed a custom neural network library that allows users to create deep learning models with an arbitrary number of layers and nodes, helping to understand the core components of neural networks...

• Implemented key neural network layers, including Linear, Sigmoid, ReLU, and Binary Cross-Entropy Loss, each with properly defined forward and backward propagation methods.

• Designed a Sequential class to manage multiple layers efficiently, making it easy to construct, train, and evaluate deep learning models in a structured manner.

• Implemented model persistence features, allowing trained models to save and load their weights from a file, ensuring easy reuse and deployment.

• Built and trained a neural network to solve the XOR problem, testing different activation functions like Sigmoid and Tanh to analyze their impact on convergence and learning performance.

• Applied the neural network library to predict NYC taxi trip durations, involving dataset preprocessing, feature engineering, and experimenting with different input transformations to improve accuracy.

• Performed hyperparameter tuning and model selection, testing different network architectures and activation functions while using early stopping to prevent overfitting.

• Benchmarked model performance against a reference model, achieving competitive RMSLE scores by selecting the most relevant features and applying proper normalization techniques.