shouldibuygastoday

Model

Gulf SST Storm Risk Model

Predicts whether to buy gas based on Gulf of Mexico sea surface temperature patterns that signal hurricane and refinery disruption risk.

Accuracy94%
AlgorithmGradientBoostingClassifier · 150 trees · depth 4 · lr 0.1
DataNOAA ERDDAP — OISST v2.1 (Optimum Interpolation Sea Surface Temperature)

Decision surface

The model's buy probability across its two most influential features. Every point was computed by calling model.predict_proba on the actual .joblib. Drag to rotate.

What the model pays attention to

sst_anomaly71%
sst_trend_7d18%
sst_7d_avg4%
sst4%
sst_30d_avg2%
hurricane_season2%

Approach

I used 2 years of daily sea surface temperature data from NOAA's OISST dataset, spatially averaged across the Gulf of Mexico (18-30°N, 80-98°W). The causal hypothesis is that elevated Gulf SSTs fuel stronger hurricanes, which shut down Gulf Coast refineries — responsible for ~50% of US refining capacity — causing gas price spikes at the pump. Features include current SST, 7-day and 30-day rolling averages, the SST anomaly (deviation from the 30-day mean), the 7-day temperature trend, and a binary hurricane season flag (June–November). I chose GradientBoostingClassifier after comparing it against logistic regression and random forest via 5-fold cross-validation. The interaction between SST anomaly and hurricane season was the strongest signal — warm anomalies outside of storm season had little predictive power, which supports the underlying causal story.

Features

sstfloat
Gulf of Mexico average sea surface temperature (°C)
sst_7d_avgderived
7-day rolling average SST
sst_30d_avgderived
30-day rolling average SST
sst_anomalyderived
SST deviation from 30-day average (°C)
sst_trend_7dderived
SST change over last 7 days (°C)
hurricane_seasonderived
Binary flag: 1 if June 1 - November 30, else 0

Built by

Cameron AnthonyLinkedIn

I have a background in sales, operations, and customer-facing leadership across tech, healthcare staffing, and mobility. I’m especially interested in AI, business strategy, and process improvement, and I enjoy finding practical ways to use technology to make teams more effective and help businesses run smarter.

Trained on

2024-01-01 to 2025-12-31

All models