Power‐Law‐Based Synthetic Minority Oversampling Technique on Imbalanced Serum Surface‐Enhanced Raman Spectroscopy Data for Cancer Screening

Surface-enhanced Raman spectroscopy (SERS) has shown highly promising for existing cancer screening. However, previous “proof-of-concept” studies ignored the natural imbalance of types in population, leading model to be biased toward learning more features majority class during process at expense ignoring minority class. Herein, a power-law-based synthetic oversampling technique (PL-SMOTE) method is proposed guide resampling multiclass serum SERS data by analyzing long-tailed (power-law) distribution prevalence population. The PL-SMOTE balances number minorities resample and overlaps between classes introducing modulating factor. Modeling on resampled datasets synthesized verifies effectiveness method. After further fine-tuning, parameters deep neural network method, an optimal screening with macroaveraged Recall score 97.24% F2-Score 97.38% obtained. A new imbalanced provided, which significant improvement performance terms also inspires other scenario, such as biological medicine, abnormal detection, disaster prediction.