This project implements a custom-built feedforward neural network (with support for stochastic or batch gradient descent) to predict flight pitch using real-world aviation data. No machine learning libraries like TensorFlow or PyTorch are used — everything is implemented using pure NumPy and Python.

The model uses a cleaned and preprocessed dataset named flight_12477, with features such as:

• altitude
• indicated_airspeed
• pitch_t-1
• roll

The target variable is the next pitch value (pitch_t), derived by shifting pitch_t-1 forward.

• Input Layer: 4 features
• Hidden Layer 1: Configurable (default: 4 neurons, ReLU)
• Hidden Layer 2: Configurable (default: 2 neurons, ReLU)
• Output Layer: 1 neuron (Linear)

Supports:

• Batch Gradient Descent
• Stochastic Gradient Descent
• Early Stopping
• Mean Squared Error and Mean Absolute Error tracking
• Gradient norm and activation plotting
• Parameter saving and logging

Make sure to install dependencies:

  pip install -r requirements.txt

  python3 NNModelFlight.py

  python3 NNModelFlight.py --epochs 500 --learning_rate 0.01 --gradient_descent stochastic --hidden_size1 4 --hidden_size2 2

  python3 NNModelFlight.py --target pitch

  python3 NNModelFlight.py --target altitude

  python3 NNModelFlight.py --target indicated_airspeed

  python3 NNModelFlight.py --target roll