The start of a new football season is the most challenging period for AI prediction models. Transfer activity reshapes squads, new managers implement different tactical systems, promoted teams are unknown quantities at their new level, and there is no current-season form data to analyze. At 1X2.TV, we have developed specific strategies for handling this transitional period.

The Cold Start Problem

In machine learning terminology, the beginning of a season represents a cold start problem. Our models rely heavily on form data — rolling averages of recent results and performance metrics — but at the start of a season, there are no recent competitive results to analyze. We address this by carrying forward a decayed version of the previous season's ratings, gradually phasing in new-season data as matches are played.

Transfer Impact Modeling

Summer transfer windows can dramatically reshape teams. Our models estimate the impact of player departures and arrivals using individual player ratings derived from their historical performance data. When a team loses its top scorer and signs an unproven replacement, our pre-season strength rating adjusts accordingly.

Promoted Team Assessment

Newly promoted teams present a particular challenge because they have no recent top-flight data. Our models use historical promotion performance data — how newly promoted teams typically perform in their first season — combined with the specific team's lower-division performance to generate initial ratings. These ratings are updated rapidly as early-season results provide new information.

Early-Season Calibration

We recommend treating early-season predictions (Matchdays 1-5) with appropriate caution. Our models explicitly communicate lower confidence levels during this period, and prediction accuracy typically improves significantly after the first month of competition as current-season data accumulates.

Tracking Pre-Season Data

1X2.TV publishes pre-season team assessments and opening predictions before every major league season kicks off, providing users with our models' initial expectations for the campaign ahead.