Personalized Cancer Diagnosis

Project Overview

This project presents an automated machine learning approach to personalized medicine. The model parses clinical genetics literature texts to map genomic variations and mutations to one of nine target clinical classes, enabling pathologists to prioritize research.

Key Features & Objectives

• Scientific Literature Parsing: Reads and handles dense medical texts containing specialized gene terminologies.
• Multi-Class Genomic Mapping: Maps combined text structures to exact target diagnosis classes.
• High-Dimensional Feature Matrices: Scales features across text columns to train predictive parameters.

Technical Stack & Technologies

• Programming Language: Python
• NLP Tools: NLTK, RegEx preprocessors
• Machine Learning: Scikit-learn, SGDClassifier (Stochastic Gradient Descent)
• Data Wrangling: Pandas, NumPy

Core Techniques & ML Methodologies

• Medical Text Vectorization: Implemented TF-IDF (Term Frequency-Inverse Document Frequency) tokenizers to convert scientific research documents into high-dimensional numerical feature blocks.
• Log-Loss Optimization: Configured multi-class Logistic Regression algorithms using log-loss calculations as the primary evaluation metric.
• Stochastic Gradient Descent (SGD): Utilized linear models optimized via SGD to handle text matrices efficiently, resolving training bottleneck steps.