Project Overview
In the healthcare industry, early detection of breast cancer significantly improves patient outcomes and reduces treatment costs. However, traditional diagnostic methods can be time-intensive and prone to error.
This project involved building a predictive analytics solution using advanced statistical modeling and machine learning techniques in R to assist clinicians and researchers with accurate and timely cancer detection.
Business Challenge
Hospitals and medical research institutions face a critical challenge:
✔ Delayed or inaccurate diagnoses can lead to higher treatment costs and increased mortality risk.
✔ There is a growing need for data-driven diagnostic tools that improve accuracy and reduce reliance on manual interpretation.
Our Solution
We developed a predictive modeling framework that combines logistic regression, discriminant analysis, and unsupervised clustering to provide accurate classification of breast cancer cases.
Key steps included:
✔ Data Preprocessing & Feature Engineering – Reduced multicollinearity and selected relevant features.
✔ Model Development:
Logistic Regression (All Features)
Logistic Regression (Secondary Features for optimized accuracy)
✔ Model Comparison: Evaluated Logistic Regression vs QDA for performance.
✔ Exploratory Analysis: Applied K-Means clustering on gene expression data to identify cancer subtypes for genomic research.
Key Results
✔ Logistic Regression (Secondary Features): 93% accuracy, Kappa = 0.84
✔ Logistic Regression (All Features): 88% accuracy, Kappa = 0.72
✔ QDA: Lower accuracy, not suitable for this dataset
✔ Gene Clustering: 4 clusters identified, aiding leukemia subtype analysis
Business Impact
For the healthcare industry, this predictive solution delivers:
✔ Improved diagnostic accuracy, reducing false positives and negatives
✔ Faster clinical decisions, improving patient care and survival rates
✔ Personalized treatment planning based on predictive insights
✔ Research advancement, supporting genomic analysis and drug development
Technologies Used
Language: R
Packages: caret, MASS, ggplot2, cluster, factoextra
Techniques: Logistic Regression, QDA, K-Means Clustering, Feature Selection