Prizma Analytics

Full breakdown on the channel 👇 youtu.be/4gmUv5SV7Aw?si…

A defensive metric with a fixed baseline isn’t fair. DDI averages two things: — the opponent’s attacking xG — the team’s xG allowed That single average becomes a dynamic baseline that adjusts for both who you’re facing and how good your defense already is.

Full breakdown on the channel 👇 youtu.be/4gmUv5SV7Aw?si…

Raw xG numbers aren’t fair to defenders. A player who played 45 minutes is being judged against one who played 90. DDI fixes that by scaling expected xG by minutes played ÷ 90 — the theoretical xG your opponent could’ve had while that defender was actually on the pitch.

Explore the DDI app yourself and see how your team’s defense ranks 👇 ddi-app.streamlit.app (heads up — if the link doesn’t load right, copy and open in your browser)

Want the full breakdown — the math, the methodology, and which teams actually rank at the top? Full video on the channel 👇 youtu.be/4gmUv5SV7Aw?si…

Defenses don’t just “hold the line” — they disrupt. DDI measures that gap: Expected xG − Actual xG When it’s positive, the defense didn’t just survive the attack. It broke it. Most metrics count goals. DDI counts damage prevented 🛡️⚽

Full methodology + walkthrough of interesting results from the app in this video youtu.be/4gmUv5SV7Aw?si…

No Opta Needed! I built my own defensive metric using public data. DDI measures how much a player suppresses realized opposition xG vs. pre-match expectations — plus an interactive app to explore it yourself. Check it out below 👇 ddi-app.streamlit.app

I built my own defensive metric using public data. No Opta needed! DDI measures how much a player suppresses realized opposition xG vs. pre-match expectations. Video explaining it below. youtu.be/4gmUv5SV7Aw?si…

If you’re interested in how K-means works to group players like this, I break it down step by step in the full video 👇 youtu.be/Ig6Gw3JB468?si…

K-means clustering groups players by playing style.
This cluster highlights wingers elite at box attacking and finishing — a powerful profile for uncovering hidden talent.

If you’re interested in how K-means works to group players like this, I break it down step by step in the full video 👇 youtu.be/Ig6Gw3JB468?si…

K-means clustering groups players by playing style. This cluster highlights wingers elite at progression and crossing — a profile that can be used to uncover hidden talent in smaller leagues.

If you’re curious how PCA actually does this, I break it down visually in this video 👇 youtu.be/Ig6Gw3JB468?si…

Most player charts compare just two raw metrics. PCA removes that constraint by rotating the data into PC space—so each axis represents many metrics at once.

If you’re interested in data science techniques like PCA and clustering, explained through football and wingers, the full breakdown is here 👇 youtu.be/Ig6Gw3JB468?si…

Data science is more fun with football — not fish or flowers. Here’s how K-means clustering classifies wingers using xG & xAG.

Full 16-min breakdown on winger PCA + clustering here youtu.be/Ig6Gw3JB468?si…

PC1 in PCA is the direction that maximizes variance. When you project players onto that line, you’re already separating how they contribute — before any clustering 👇