HammerPricer

Insight Data Science Fellowship demo project (September 2014).

Transcript

1. HammerPricer Art Auction Pricing Jonathan Whitmore
2. The Problem Market over $60 Billion per year Opaque pricing Hard to value as short or long-term investment Deux personnages (1953) By: Karel Appel
3. The Problem Market over $60 Billion per year Opaque pricing Hard to value as short or long-term investment Deux personnages (1953) By: Karel Appel Can I predict the “hammer price” for art sold at auction?
4. DEMO
5. Algorithm Algorithm(s) python: scikit-learn’s linear regression. Most important features 1. Auction house estimates 2. Artist 3. Auction house sold 4. Creation Date 5. Physical Dimensions
6. Feature Importance – SelectKBest 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Mean Squared Predicted Error + Color Analysis + Country of Sale + Physical Dimensions + Creation/Sale Date + Auction House + Artist Auction Estimate
7. HammerPricer Linear Regression Results $400 $3,000 $22,000 $162,800 $1,202,600 $8,886,100 $65,660,000 Actual Hammer Price [$] $400 $3,000 $22,000 $162,800 $1,202,600 $8,886,100 $65,660,000 Predicted [$] Training (5677) Testing (430) MSE Training 0.395 MSPE Testing 0.392
8. Past year absolute error in prediction Auction HammerPricer 0 5 10 15 20 25 Cumulative Error [$] (in millions) $4.35 million saved
9. Jonathan Whitmore PhD Physics UC San Diego Postdoc at Swinburne Co-starred in 3D IMAX film