Weighted F1 Measure from Group Counts Method

A Weighted F1 Measure from Group Counts Method is a weighted-average performance measure computation method that computes weighted means of per-class F1 scores using class support weights or custom importance weights.

• AKA: Support-Weighted F1 Method, Class-Weighted F1 Computation, Importance-Weighted F1 Method, Custom-Weighted F1 Average Method.
• Context:
  • It can typically compute Weighted-F1 = Σ(w_i * F1_i) / Σ(w_i) for class weights w_i.
  • It can typically use class support sizes as weights: w_i = n_i / N_total.
  • It can typically balance between Macro-F1 Measure from Group Counts Method and Micro-F1 Measure from Group Counts Method.
  • It can often incorporate domain-specific importance through custom weights.
  • It can often handle class imbalance better than unweighted macro-average.
  • It can often be reported by default in scikit-learn classification report.
  • It can range from being a Support-Weighted F1 Measure from Group Counts Method to being a Custom-Weighted F1 Measure from Group Counts Method, depending on its weight source.
  • It can range from being a Proportional-Weighted F1 Measure from Group Counts Method to being a Binary-Weighted F1 Measure from Group Counts Method, depending on its weight distribution.
  • It can range from being a Static-Weighted F1 Measure from Group Counts Method to being a Adaptive-Weighted F1 Measure from Group Counts Method, depending on its weight adjustment.
  • It can range from being a Normalized-Weighted F1 Measure from Group Counts Method to being an Unnormalized-Weighted F1 Measure from Group Counts Method, depending on its weight scaling.
  • ...
• Example(s):
  • Support-Weighted F1 Calculations, such as:
    • Class A (n=500): F1=0.85, w=0.5; Class B (n=300): F1=0.75, w=0.3; Class C (n=200): F1=0.90, w=0.2; Weighted-F1 = 0.5*0.85 + 0.3*0.75 + 0.2*0.90 = 0.83.
    • Natural class frequency weighting in imbalanced datasets.
  • Cost-Weighted F1 Calculations, such as:
    • Medical diagnosis: critical conditions weighted 10x higher.
    • Fraud detection: high-value transactions weighted more.
  • Business-Importance Weighted F1s, such as:
    • Customer segments weighted by revenue contribution.
    • Product categories weighted by strategic importance.
  • ...
• Counter-Example(s):
  • Macro-F1 Measure from Group Counts Method, which uses equal weights.
  • Micro-F1 Measure from Group Counts Method, which pools counts instead.
  • Median F1 Method, which uses order statistics.
• See: Performance Measure Computation Method, Weighted Average, F1 Score, Class Weight, Support Size, Macro-F1 Measure from Group Counts Method, Micro-F1 Measure from Group Counts Method, Class Imbalance Problem, Cost-Sensitive Learning, Multi-Class Classification Evaluation.

---