Cancer-Diagonsis

Personalized Cancer Diagnosis A machine learning project focused on building models for personalized cancer treatment prediction using genetic mutation and textual data. This repository contains modular Python scripts for preprocessing, feature engineering, training classifiers, and evaluating performance metrics.

Models Implemented

Model	File	Description
K-Nearest Neighbors	cancerKNN.py	Distance-based classification using KNN
Logistic Regression	cancerLogisticReg.py	Linear model using log loss
Naive Bayes	cancerNB_Model.py	Probabilistic classifier with calibration
Support Vector Machine	cancerSVM.py	Margin-based classifier
Random Forest	cancerRandomForest.py	Tree ensemble model
Stacking Classifier	cancerStackingClassifier.p`	Combines multiple models using meta-classifier
Majority Baseline	cancerMajorityClassifier.p`	Always predicts most frequent class
Random Model	cancerRandomModel.py	Random prediction baseline

Feature Engineering

Script	Feature Used	Description
cancerUnivariateText.py	Text-based features	Extracted using TF-IDF or CountVectorizer
cancerUnivariateGene.py	Gene features	One-hot or response encoding
cancerUnivariateVariation.py	Variation features	One-hot or response encoding

Requirements: pip install -r requirements.txt

Typical Libraries: numpy,pandas,scikit-learn,matplotlib,seaborn,nltk

How to Run

1. Clone the repository: git clone https://github.com/Thevishal-kumar/Cancer-Diagonsis.git cd personalised-cancer-diagnosis

2. Run individual model scripts: python cancerLogisticReg.py

3. Each script will: Load and preprocess the data Train the model Calibrate the classifier (if needed) Evaluate using Log Loss and Confusion Matrix

Evaluation Metrics: Log Loss Confusion Matrix Precision, Recall, F1-Score t-SNE Visualization for class separation

Dataset: This project uses data provided by the MSK-IMPACT Kaggle competition. It includes: Patient gene and mutation information Variation descriptions Text evidence (clinical notes)