🫁 Lung Cancer Risk Analysis Platform

---

Interactive Machine Learning Platform for Lung Cancer Risk Assessment
Empowering healthcare professionals with AI-driven insights for early detection and prevention.

🌐 Live Demo · 🐛 Report Bug · 💡 Request Feature

---

---

📋 Table of Contents

• Overview
• Dataset Description
• Technical Challenges
• Exploratory Data Analysis
• Key Findings
• Dashboard Features
• Installation
• Project Structure
• Technologies
• Future Enhancements
• Contributing
• License
• Contact

---

🎯 Overview

This project delivers a full data science pipeline for lung cancer risk prediction, integrating statistical analysis, machine learning, and interactive dashboards built with Streamlit.

Healthcare professionals can analyze clinical risk factors, visualize data patterns, and evaluate model predictions through a user-friendly web interface.

Core Objectives

• 🔍 Identify significant behavioral and clinical risk factors
• 📊 Develop and validate robust machine learning models
• 🎨 Build an interactive platform for data exploration
• 📈 Generate actionable insights for early detection

---

📊 Dataset Description

Source: Clinical survey (survey lung cancer.csv)
Samples: 309 Features: 16 (15 predictors + 1 target)
Missing Values: None
Class Imbalance: 87% cancer, 13% healthy

Category	Variables	Type	Description
Demographic	AGE, GENDER	Continuous, Binary	Age (21–87), sex
Behavioral	SMOKING, ALCOHOL_CONSUMING, PEER_PRESSURE	Binary	Lifestyle & exposure
Symptoms	YELLOW_FINGERS, ANXIETY, FATIGUE, ALLERGY, WHEEZING, COUGHING, SHORTNESS_OF_BREATH, SWALLOWING_DIFFICULTY, CHEST_PAIN	Binary	Clinical symptoms
Comorbidities	CHRONIC_DISEASE	Binary	Pre-existing conditions
Target	LUNG_CANCER	Binary	Diagnosis (YES/NO)

Encoding: 1 → NO, 2 → YES

---

⚙️ Technical Challenges

1️⃣ Data Preprocessing

Non-standard binary encoding required conversion:

binary_cols = [...]
df[binary_cols] = df[binary_cols].replace({1: 0, 2: 1}).astype('uint8')

✅ Result: Uniform encoding and reduced memory footprint.

---

2️⃣ Class Imbalance

Problem: 87:13 imbalance → biased predictions.

---

3️⃣ Feature Scaling

from sklearn.preprocessing import StandardScaler
df['age_scaled'] = StandardScaler().fit_transform(df[['age']])

---

✅ Result: Improved convergence for distance-based models (SVM, Logistic Regression).

---

🔬 Exploratory Data Analysis

🧓 Age Differences

• Cancer: 62.95 ± 7.97
• Healthy: 60.74 ± 9.63
• t-test: p > 0.05 → Age is not a significant predictor

---

🔝 Top 5 Predictive Symptoms

Symptom	Healthy (%)	Cancer (%)	Odds Ratio	p-value
Allergy	12.8	61.9	11.03	< 0.001
Alcohol consume	17.9	61.1	7.18	< 0.001
Swallowing difficulty	12.8	51.9	7.32	< 0.001
Wheezing	23.1	60.4	5.06	< 0.001
Coughing	25.6	62.6	4.85	< 0.001

---

🔗 Correlations

• Anxiety ↔ Yellow Fingers: r = 0.56
• Anxiety ↔ swallowing_difficulty: r = 0.48
• Gender ↔ Alcohol consume: r = 0.45

---

🧩 Clustering Analysis

Unsupervised K-Means (k=3) revealed distinct patient profiles:

• Cluster 0: Smoking men with yellow fingers and fatigue, but no alcohol consumption
• Cluster 1: Wpmen with respiratory symptoms and alcohol consumption.
• Cluster 2: Men and women with yellow fingers and difficulty swallowing.

---

🎯 Key Findings

• 10 binary symptoms significantly associated (p < 0.05)
• Top Model: Gradient Boosting

Metric	Score	Interpretation
ROC-AUC	0.959	Excellent discrimination
Accuracy	91%	High correctness
Precision	96.9%	Few false positives
Recall	92.6%	Few missed cases
F1-Score	0.947	Balanced performance

---

🖥️ Dashboard Features

Page	Description
🏠 Home	Summary KPIs & dataset overview
📊 Distribution	Histograms, boxplots, descriptive stats
🔄 Correlations	Heatmaps and pairwise associations
📈 Statistical Tests	t-tests, Chi-square, Mann-Whitney U
🎯 Clusters	3D PCA + radar charts
🧬 Prediction	ML-based risk calculator
🎓 Model Performance	ROC curves, calibration, confusion matrix

---

🎨 Interactive visuals include:

• 3D PCA plots
• Feature importance bars
• Radar charts
• Calibration curves

---

🚀 Installation

# 1. Clone repository
git clone https://github.com/aespinosa95/lung_cancer.git
cd lung_cancer

# 2. Create virtual environment
python -m venv lungcancer
source lungcancer/bin/activate  # On Windows: lungcancer\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Launch the Streamlit dashboard
streamlit run src/Home.py

📁 Project Structure

lung_cancer/
├── data/
│   ├── survey lung cancer.csv
│   └── survey lung cancer_clean.csv
├── notebooks/
│   └── preprocess.ipynb
│	└── EDA.ipynb
├── src/
│   ├── Home.py
│   ├── pages/
│   └── utils/
├── requirements.txt
├── README.md
└── LICENSE

---

🛠️ Technologies

Tool	Purpose
Python 3.12+	Core language
Streamlit	Dashboard framework
Pandas / NumPy	Data handling
Scikit-learn	ML algorithms
Plotly / Matplotlib / Seaborn	Visualization
SciPy	Statistical testing

---

Areas for Contribution

• 🧠 ML model improvements
• 📊 New visualizations
• 🧩 Code optimization
• 🧾 Documentation enhancement

---

📄 License

Distributed under the MIT License.
See the LICENSE file for details.

---

📬 Contact

• 🐙 GitHub: @aespinosa95
• 💼 LinkedIn: Asunción Espinosa Sánchez
• 🌐 Portfolio: asuncionespinosa
• 📁 Project Link: lung_cancer repo

---

⚕️ Disclaimer:
This platform is for research and educational purposes only.
Predictions should be interpreted by qualified healthcare professionals.

Made with using Streamlit and Python.