Cancer Prediction using MLP and CNN

This project implements two machine learning approaches for cancer prediction:

1. A Multi-Layer Perceptron (MLP) for lung cancer risk prediction
2. A Convolutional Neural Network (CNN) for skin cancer type classification

Project Structure

.
├── MLP/                    # Lung Cancer Prediction
│   ├── Cancer_application.R    # Shiny application for visualization
│   ├── Lung_Cancer_Dataset.csv # Dataset with 5,000 records
│   ├── NN_Lung_Cancer.Rmd      # MLP implementation
│   └── NN_Lung_Cancer_NotPrebuilt.Rmd
└── CNN/                    # Skin Cancer Classification
    ├── CNN.ipynb           # CNN implementation
    └── accuracy_plot.png   # Model performance visualization

MLP for Lung Cancer Prediction

The MLP model predicts lung cancer likelihood based on 18 risk factors from a dataset of 5,000 records. The model features:

• 5 hidden layers
• Input features related to lung cancer risk factors
• Interactive visualization through a Shiny application

Running the MLP Application

1. Open MLP/Cancer_application.R in R Studio
2. Click the "Run App" button (green arrow) in the top right corner
3. The application will launch on localhost (typically takes ~5 minutes to load)

CNN for Skin Cancer Classification

The CNN model classifies skin cancer types using the HAM10000 dataset. The implementation includes:

• Convolutional neural network architecture
• Performance visualization
• Jupyter notebook with detailed implementation

Running the CNN Model

1. Open CNN/CNN.ipynb in Jupyter Notebook
2. Run the cells sequentially to train and evaluate the model

Requirements

MLP Requirements

• R Studio
• Required R packages:
  • shiny
  • neuralnet
  • ggplot2
  • dplyr

CNN Requirements

• Python 3.x
• Jupyter Notebook
• Required Python packages:
  • tensorflow
  • keras
  • numpy
  • pandas
  • matplotlib

Dataset Information

Lung Cancer Dataset

• 5,000 records
• 18 features related to lung cancer risk factors
• Binary classification task (cancer likelihood)

Skin Cancer Dataset (HAM10000)

• 10,000 dermatoscopic images
• 7 different types of skin cancer
• Multiclass classification task