Art Of Data

---

Project Description

Predicting the final sale price at fine art auctions

+++++++++++++++++

Overview:
With data retrieved from an art database create a model that would be capable of utilizing the numeric data to predict the price of 10 artists sale prices.
Method
Regression / Classification / Deep Learning (Image Recognition).
Takeaways
TBD.
Skills:
Tensor Flow, Python, Natural Language Processing (NLP), Pandas, EDA, Facebook, SciKit-Learn, Classification

Project Goals

---

• Accurately predicting an auction house estimate price for the art of 10 artist of which are the highest grossing/ highest volume within the market

Initial Thoughts

---

• The artist will likely be the strongest drivers of the price point. The disparity of sale price for some artists is much higher meanwhile some will float within the same range.
• When incorporating image recognition the neural network will likely be able to recognize the works of the 10 artists, due to all of them being regarded for their individual styles.

Planning

---

1. Acquire data from artnet.com
2. Data could only be retrieved in the form of a PDF
3. Clean aforementioned PDF/PDFs
4. Explore and analyze data for better insights on relative context to artist and their prices
   a. Determine our baseline prediction
   b. Are we beating the auction house's estimates?
   c. Does logarithmically transforming hammer price allow for easier scaling for models? d. Stats test high frequency words to artists\
5. Start with Regression Model
6. Proceed to Classification
7. Document conclusions, recommendations, and next steps
8. Move forward to MLP (Multilayer Percepitron) OR CNN (Convultional Neural Network)

Data Dictionary

---

Feature	Definition
artist	The name of the artist who created the artwork
name	The name of the artwork
medium	The medium used to create the artwork
size	The size of the artwork
date_created	The date the artwork was created
lot	The lot number of the artwork in the auction
date_sold	The date the artwork was sold
auction_house	The name of the auction house where the artwork was sold
edition	The edition number of the artwork
estimated_price	The estimated price of the artwork before the auction (leakage)
hammer_price	The final price of the artwork at the auction

Reproducability Requirements

---

6. ⚠️ Disruption in reproducibility - artnet.com

Terms of Service Ch.1:
1. …You may not modify, create derivative works from, participate in the transfer or sale of, post on the web, or in any way exploit the Site or Services or any portion thereof for any public or commercial use without the express written permission of artnet… 2. We are forced to comply to these terms and may not offer the data to any of those that wish to reproduce/ alter or results.

Conclusions

• Regression models can get fairly accurate when predicting the price of art
• Better usage of deep learning model as it is not accurate at all
• Artists still producing art are more predictable

Recommendation

• Provide the art industry with a better tool that will potentially allow for higher liquidity within the market and provide more transparency and visibility for consumers