UFC Fight Night: Allen vs. Costa: Predictions & Analysis
UFC Fight Night: Allen vs. Costa lands on Saturday, May 16, 2026 in Las Vegas, Nevada, USA with 1 bouts on the card. Below is our fight-by-fight breakdown, combining Elo ratings, rolling statistical trends, style matchup data, and betting market context into a pick for every bout.
Quick Picks
| Matchup | Pick | Confidence | Prob |
|---|---|---|---|
| Arnold Allen vs Melquizael CostaFeatherweight | Melquizael Costa | Lean | 57% |
Like these picks? Bet on DraftKings
Place your bets on the fights above at DraftKings Sportsbook
Bet Now on DraftKingsFight-by-Fight Breakdown
Arnold Allen vs Melquizael Costa
The Featherweight matchup features Arnold Allen (11-3-1) taking on Melquizael Costa (7-2-1).
The Elo gap here is razor-thin — Allen at 1610, Costa at 1619. On paper, this is as close to a coin flip as it gets. Costa has won 6 straight.
Both fighters land in our "Brawler" archetype — aggressive fighters who push forward, throw volume, and look to overwhelm opponents in exchanges. When mirror matchups like this happen, the edge usually goes to whoever can impose their preferred pace and range.
A few statistical edges stand out. Costa throws significantly more leather — a 1.6 sig. strike per minute gap. Costa is far more active with takedowns, averaging 1.5 more per 15 minutes. Allen has tighter striking defense, making opponents miss more often.
The Pick: Melquizael Costa over Arnold Allen. The model gives Costa a slight nod at 57% — this could easily go either way.
Methodology
Predictions are generated by our ensemble model combining LightGBM (65%) and CatBoost (35%), trained on every UFC fight since 1994. The model uses 23 features including Elo ratings, rolling 5-fight statistical averages, style matchup history, physical attributes, and market odds when available.
On our held-out test set (402 fights from January-September 2023), the model achieves 63.4% accuracy with a log-loss of 0.626. High-confidence picks (>75% probability) hit at 82.7%. For full model transparency, visit our Model page.