Data Science 18
- Five Archetypes at the Top of YouTube: Creator Segmentation on the Global Top-995
- 93,000 Customers and No Repeats: Olist Brazilian E-Commerce Analytics
- When a Naive Baseline Beats LightGBM: Bike-Sharing Demand with Proper Time-Series CV
- Geography Beats the Odometer: NYC Taxi Trip Duration Prediction
- Quality, Then Area, Then Everything Else: Ames Housing with Stacked Ensembles
- When the Description Field Does the Work: A Netflix Content Recommender
- A Spotify Recommender That Returns Rhythm-of-Love for a Death-Metal Query
- Room Type First, Geography Second: Airbnb NYC 2019 Price Modelling
- Contract Type Eats the Rest: Telco Customer Churn with SHAP
- The Default Threshold Is the Bug: Credit Card Fraud on the ULB Benchmark
- Auditing a 15,225-Image Tomato Leaf Disease Dataset Before Training a Single Model
- Ranking 528,000 FDA Adverse Event Reports by Hospitalization Risk
- The One Human a 99.84 Percent AI Text Detector Got Wrong
- A Classifier That Reads No Articles Still Hits F1 = 1.000 on This Fake News Benchmark
- Why I Picked the Less Accurate Model for Stroke Screening
- When Kawasaki Disease Started Looking Like a Weather Map
- The California Housing Dataset Is Mostly a Geography Lesson
- The Mental Model That Made Neural Networks Finally Click for Me