opeoluwa22/Lifestyle-Based-Cervical-Cancer-Risk-Prediction-
Domain:
healthcareCreator:
ope
Host:
Lifestyle-Based Cervical Cancer Risk Prediction and Population Burden Analysis in West Africa
# **Lifestyle-Based Cervical Cancer Risk Prediction and Population Burden Analysis in West Africa**
## **Objective**
This project aims to analyse lifestyle-related risk factors and predict cervical cancer risk using machine learning models, while also examining the population burden of the disease across West African countries.
## **Dataset Description**
Two main datasets were used:
* **Individual-level dataset:**
Contains demographic, behavioural, and clinical variables associated with cervical cancer risk (e.g., age, smoking habits, sexual history, contraceptive use, and STD history). The target variable is biopsy-confirmed cervical cancer.
* **Population-level dataset (GBD):**
Includes incidence and prevalence data across West African countries, enabling regional trend and burden analysis.
## **Models Used**
The following machine learning models were implemented and compared:
* Logistic Regression (baseline model)
* Random Forest (ensemble learning)
* XGBoost (gradient boosting with hyperparameter tuning)
To address class imbalance, **random oversampling** and **threshold tuning** were applied, with a focus on improving recall for positive cancer cases.
## **Key Results**
* The dataset exhibited **severe class imbalance** (very few positive cancer cases).
* **Accuracy was not a reliable metric**, as models could perform well while failing to detect positive cases.
* After applying oversampling and threshold tuning:
* **Logistic Regression (Oversampled):**
* Accuracy: ~0.78
* Recall: ~0.18
* **XGBoost (Tuned + Threshold = 0.05):**
* Accuracy: ~0.76
* Recall: ~0.55
* **Random Forest (Tuned + Threshold):**
* Accuracy: ~0.61
* **Highest Recall: ~0.82**
* **Best model for detection:** Random Forest (highest recall)
* **Best balanced performance:** XGBoost
These results highlight the trade-off between accuracy and recall in imbalanced medical datasets.
## **Key Insights**
* Clinical screening variables showed stronger predictive power than lifesty …
Similar
Ranking lifestyle risk factors for cervical cancer among Black women: A case-control study from Johannesburg, South Africa<scp>DNA</scp> methylation biomarkers for cervical cancer risk prediction in <scp>HIV</scp>‐positive Nigerian womenPerformance Evaluation of Machine Learning Models For Cervical Cancer PredictionMycoplasma Co-Infection Is Associated with Cervical Cancer RiskLifestyle and Reproductive Risk Factors Associated with Cervical Dysplasia among Women in Rivers State, NigeriaSurvival Prediction of Cervical Cancer Patients using Genetic Algorithm-Based Data Value Metric and Recurrent Neural Network