Research / applied visualization

Using interactive data visualization for DeFi market analysis

An automated multi-source ETL workflow and interactive analytical interface for examining DeFi market structure and movement.

Authors

Daria Pavlova
Inna Novalija

Venue

Information Society 2025
Slovenian KDD Conference
Ljubljana, Slovenia

Access

Reported full ETL

45 s

The paper also reports an 8-second incremental execution path.

Reported coverage

6,000+

protocols processed through the DeFiLlama source path.

Source APIs

3

CoinGecko, DeFiLlama and DexScreener.

Research question

Can an automated data pipeline and coordinated visual views reduce the friction of examining DeFi market structure and short-term movement?

The paper combines market and protocol data from three APIs, validates and transforms it into analytical outputs, and presents the result through synchronized Tableau views.

System view

Figure 01

From source data to analytical views.

01 Source APIs

Market & protocol data

CoinGecko
DeFiLlama
DexScreener

02 Data processing

Validate

Completeness, consistency, timeliness and outliers.

Transform

Normalize timestamps, derive features and aggregate.

03 Analytical delivery

Coordinated views

Overview, time series, movers and categories.

Four connected analytical outputs

The figure follows the system described in the paper: three source APIs pass through explicit validation and transformation boundaries before analytical delivery.

Method

  1. Collect token, protocol and exchange-pair data through three API clients with retry and backoff handling.
  2. Validate completeness, consistency, timeliness and outliers before transformation.
  3. Normalize timestamps, derive rolling features and aggregate market categories.
  4. Publish four analytical outputs for coordinated overview, time-series, movers and category views.

Results

MeasurementReported valueConditions stated in the paper
Full ETL execution45 sComplete multi-source pipeline
Incremental execution8 sIncremental update path
Initial dashboard load3.2 s ± 0.5n = 100
Filter operation1.8 s ± 0.3Interactive dashboard
Peak memory256 MBReported pipeline run

Full ETL execution

Reported value
45 s
Conditions stated in the paper
Complete multi-source pipeline

Incremental execution

Reported value
8 s
Conditions stated in the paper
Incremental update path

Initial dashboard load

Reported value
3.2 s ± 0.5
Conditions stated in the paper
n = 100

Filter operation

Reported value
1.8 s ± 0.3
Conditions stated in the paper
Interactive dashboard

Peak memory

Reported value
256 MB
Conditions stated in the paper
Reported pipeline run
Interpretation

These are measurements reported in the publication. A public reproducibility package is not currently linked here, so the portfolio does not present them as independently revalidated benchmarks.

Limits

  • The paper notes that rehypothecation can inflate TVL measurements by 20–30%.
  • Source APIs may lag by 5–15 minutes during high-volatility periods.
  • The reported coverage excludes protocols below $1M TVL.
  • User-efficiency and use-case claims remain paper-reported until their underlying evaluation artifacts are made public.

Other research

01 record