[The Snap] WR Breakout Model

Player mentions

Article text

First things first, if you like the post, join our growing community / free newsletter, The Snap: fantasy football news, stories, discussions, opinions, injury monitoring, and humor gems delivered multiple times a week to your inbox. Stay caught up and don't miss a thing as we approach draft season. 👉 https://thesnap.beehiiv.com/ --- What's up everybody, I've been tinkering with some rough-and-dirty ML experiments to surface WR breakout candidates for next season. Posting because the results for this most recent season (below) are quite strong. I'd also like to solicit feedback on improvements and ideas to the features being used below. The goal: rank year-2/3/4 WRs by predicted probability of breaking out next season. I somewhat arbitrarily defined "Breakout" is defined as either: next-season PPR PPG ≥ prior + 3 AND next ≥ 14 PPG next-season PPR PPG ≥ prior + 4 AND next ≥ 12 PPG The OR rule catches two archetypes: the "good-guy-gets-great" jump (a 11→14 PPG year) and the "low-base-explosion" jump (a 8→12 PPG year). Year 2/3/4 only because that's the typical breakout window — rookies are too noisy, year-5+ guys are usually settled into their ceiling. Training data: three season-pairs — 2022→2023, 2023→2024, 2024→2025. About 318 total year-2/3/4 candidates across the three seasons, but only 14 of them are positives. I used a leave-one-season-out cross-validation approach and prevent the. Hopefully going to widen this dataset (main limitation is in getting the YPRR numbers) but the early results are still interesting. Features (8 total): Draft capital — piecewise score by overall pick. R1 picks → [.8,1], R2 → [.6,.8], R3 → [.4,.6], R4-7 → [0,.4], undrafted = 0. YPRR (raw) — yards per route run from the prior season. Drop players with fewer than 50 routes. Available targets — total vacated targets for the player's projected next-season team (targets that previously went to WRs/TEs/RBs no longer on the roster, accounting for trades and free agency). YPRR × Available targets — interaction term capturing efficiency multiplied by opportunity. Top-2 game PPR sum — sum of the player's two highest single-game PPR totals from the prior season. Catches flash upside. Career games — total regular-season games played up through the prior season. Age — player's age at September 1 of next season. Prior PPG — prior-season PPR PPG Class balance is severe (no getting around the fact that in reality breakout candidates are low) so predicted probabilities skew low. Treat them as a ranking signal, not calibrated frequencies. Last season (2025's) breakout candidates: https://preview.redd.it/cakxepa44r1h1.png?width=1032&format=png&auto=webp&s=16dae80d8697ebcde55ee9d68dfe360ecb624c0e player team pos year draft_capital yprr_raw available_targets_raw breakout_top2_raw career_games age yprr_x_at prior_ppg breakout_prob Drake London ATL WR 4 0.9548 2.31 2 74.1 50 24.1068 4.62 16.5176 0.4588 Jaxon Smith-Njigba SEA WR 3 0.8774 1.81 263 61 34 23.5455 476.03 14.8824 0.0755 Rome Odunze CHI WR 2 0.9484 1.17 141 43.6 17 23.2471 164.97 8.5235 0.06 Chris Olave NO WR 4 0.9355 2.12 139 39.3 39 25.18 294.68 9.5875 0.0446 Garrett Wilson NYJ WR 4 0.9419 1.68 221 59.1 51 25.1116 371.28 14.8176 0.0136 Marvin Harrison Jr. ARI WR 2 0.9806 1.63 14 52.1 17 23.0582 22.82 11.5588 0.0115 Parker Washington JAX WR 3 0.1228 1.02 221 36.4 22 23.4497 225.42 6.9286 0.0084 Puka Nacua LA WR 3 0.148 3.55 220 67.1 28 24.2601 781 18.7818 0.0082 Marvin Mims Jr. DEN WR 3 0.6065 2.58 208 52.3 33 23.4…