🩺 Cancer Data Analysis (2015–2024)

📌 Overview

This project explores a global cancer patient dataset (2015–2024) to understand:

• Patient demographics
• Lifestyle & environmental risk factors
• Treatment effectiveness
• Survival outcomes

We combine exploratory data analysis (EDA), statistical hypothesis testing, and machine learning modeling to extract insights that can support healthcare decision-making, clinical strategy, and future predictive systems.

---

🎯 Objectives

• Identify key risk factors influencing cancer survival.
• Study demographic patterns (age, gender, country).
• Evaluate treatment effectiveness across chemotherapy, surgery, radiation, and combinations.
• Test hypotheses regarding stage, treatment, and geography.
• Build initial predictive models for survival outcomes.
• Provide policy-level and clinical recommendations.

---

📊 Dataset

• Source: Global Cancer Patients Dataset (2015–2024)
• Observations: Thousands of patient-level entries

Features:

• Demographics: Age, Gender, Country, Year of Diagnosis
• Risk Factors (scaled 0–10): Genetic Risk, Smoking, Alcohol Use, Obesity, Air Pollution
• Clinical: Cancer Stage, Treatment Type
• Outcome: Survival Status, Survival Months

Data Quality:

• ✅ No missing values
• ✅ No inconsistencies
• ✅ Risk factors standardized

---

🔍 Exploratory Data Analysis (EDA)

👥 Demographics

• Age: 20–89 years (mean ~54). Risk rises sharply after 50.
• Gender: Balanced dataset.
• Year of Diagnosis: 2015–2024 (median ~2019).

⚠️ Risk Factors

• All risk scores follow symmetric distributions (mean ~5, std ~2.8).
• Normalized → fair comparisons across patients.

🧬 Cancer Stage

• Most patients diagnosed at Stage II & III.
• Stage IV shows lowest survival months.

💊 Treatment Effectiveness

• Treatments: Chemotherapy, Surgery, Radiation, Combinations.
• Combination therapy (Chemo + Surgery) → highest survival.
• Single treatments less effective.

🌍 Geographic Variation

• Developed countries: Higher survival due to early detection & advanced care.
• Developing countries: Late-stage diagnoses, shorter survival.

---

📑 Hypothesis Testing

Hypothesis 1: Cancer Stage vs Survival

• H0: Survival independent of stage
• H1: Survival differs across stages
• Result: p < 0.05 → Reject H0
  ✅ Survival significantly affected by cancer stage

Hypothesis 2: Treatment Type vs Survival

• H0: Survival independent of treatment
• H1: Survival differs across treatments
• Result: p < 0.05 → Reject H0
  ✅ Combination therapies improve survival outcomes

Hypothesis 3: Country vs Survival

• Result: p > 0.05 → Fail to reject H0
  ⚠️ Differences exist descriptively, but not statistically significant
  → Additional socio-economic data needed

---

🤖 Machine Learning Models

Models Tested

• Logistic Regression (LR)
• Stochastic Gradient Descent (SGD)
• Gradient Boosting (GB)
• Random Forest (RF)

Observations

• All models initially showed 100% accuracy
• ⚠️ Indicates overfitting / data leakage
• Likely due to strong correlation between Stage and Survival

Next Steps

• Apply train-test split with stratification
• Use cross-validation
• Handle class imbalance (SMOTE, undersampling)
• Engineer interaction features (e.g., Stage × Treatment)

---

💡 Key Insights

1. Risk increases after age 50 → target screening
2. Cancer stage at diagnosis is the most critical survival factor
3. Combination therapies (Chemo + Surgery) extend survival significantly
4. Geographic disparities highlight healthcare inequality
5. Predictive modeling needs refinement to prevent overfitting

---

🛠️ Tech Stack

• Languages: Python
• Libraries: Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, Statsmodels
• Methods:
  • Exploratory Data Analysis (EDA)
  • Hypothesis Testing (ANOVA)
  • Predictive Modeling (LR, RF, GB)

---

🚀 How to Run

Clone the repository

git clone https://github.com/your-username/cancer-data-analysis.git cd cancer-data-analysis

Install dependencies

pip install -r requirements.txt

Run Jupyter Notebook

jupyter notebook Cancer_Data_Analysis.ipynb

For Data Science

Refine models with cross-validation

Perform feature engineering for better predictive power

Explore survival analysis (Kaplan-Meier, Cox Regression)

Deploy results in a dashboard or API

⚠️ Limitations

No socio-economic variables included

Risk of model overfitting

Dataset does not separate cancer subtypes

Limited post-treatment follow-up data

🔮 Future Work

Apply time-to-event survival models

Add socio-economic & healthcare access data

Build an AI-powered cancer outcome predictor

Deploy findings into a web-based clinical dashboard

👨‍💻 Contributors

Bhoopendra Vishwakarma – Data Analyst | Python • SQL • Power BI • Machine Learning linkedin

Yogesh Chouhan – Collaborator & Contributor ( Data Scientist ) linkedin