Barbados Traffic Analysis Challenge
ML solution for predicting traffic congestion using video analysis. Implemented YOLO detection, VideoMAE embeddings, and ensemble models.
Bioinformatics · Machine Learning · Clustering · Healthcare AI
Python, Scikit-learn, PCA, t-SNE
★ 2 stars on GitHub LinksCode All projects
Overview
Application of unsupervised clustering algorithms to RNA-Seq gene expression data from the TCGA Pan-Cancer project. The goal is to automatically group cancer samples into clusters that reveal functional or phenotypic similarities without using tumor labels.
Dataset
TCGA Pan-Cancer RNA-Seq (HiSeq)
- Samples: 881 patients
- Features: 20,531 genes
- Classes: 5 tumor types (BRCA, KIRC, COAD, LUAD, PRAD)
- Source: UCI ML Repository
Methodology
Dimensionality Reduction
- PCA: Principal Component Analysis for initial reduction
- t-SNE: For 2D visualization of high-dimensional clusters
Clustering Algorithms
K-Means
- Optimizes intra-cluster inertia
- Elbow method for optimal K selection
- Well-suited for spherical clusters
DBSCAN
- Density-based clustering
- Detects noise and outliers
- Handles arbitrary cluster shapes
Evaluation Metrics
- Silhouette Score
- Elbow Method
- Visual cluster separation analysis
Key Findings
The unsupervised approach successfully identified distinct gene expression patterns corresponding to different tumor types, demonstrating the potential for:
- Automatic tumor classification
- Discovery of cancer subtypes
- Biomarker identification
Applications
- Clinical Diagnosis: Automated tumor type identification
- Drug Discovery: Identifying similar tumor profiles for treatment
- Research: Understanding gene expression patterns in cancer