AI Without “Black Boxes”
Unlike using high-level libraries, this project stands out by manually implementing the learning algorithms. I developed a fully functional neural network using only linear algebra and calculus, applied to the critical classification of medical tumors.
Software Engineering & Machine Learning
- Neural Network from Scratch: Implementation of the Backpropagation algorithm and activation functions using purely NumPy, demonstrating a deep understanding of Deep Learning fundamentals.
- Dynamic Training: The model allows real-time adjustment of hyperparameters (learning rate, epochs, hidden neurons, and cost functions) directly from the UI to experiment with model convergence.
- Pythonic Architecture: Clean and robust code leveraging Type Hinting and the
attrsmodule for professional-grade data management. - Multi-paradigm Design: Structured approach combining Object-Oriented Programming (OOP) for the network architecture and Functional Programming for data processing.
Scientific Analysis & Visualization
- Control Dashboard: A graphical interface built with Tkinter that enables Exploratory Data Analysis (EDA) by selecting variables to identify key correlations.
- Real-time Monitoring: Interactive visualization of the decision boundary and the error curve during training via Matplotlib integration, showcasing how the AI “learns” to separate data points.
- Industry Dataset: Leveraged the Scikit-learn breast cancer dataset to validate model accuracy against industry-standard benchmarks.