Robust and automated analysis framework for photoelectron yield spectroscopy: overcoming statistical bias via MAE-based AIC and log-log linearization

We propose an integrated analysis framework to objectively and accurately determine the ionization potential (Ith) and the power-law exponent (n) from photoelectron yield spectroscopy (PYS) spectra. Conventional analysis methods relying on linearized plots are fundamentally susceptible to statistical bias caused by data transformation, which often distorts noise distribution and leads to model misclassification. To remove this bias, we employ the Akaike Information Criterion (AIC) based on the Mean Absolute Error (MAE) calculated in the original physical space. This approach makes the evaluation robust against outliers and allows objective comparisons between physically meaningful fixed-exponent models (e.g. n = 2, 3). Furthermore, we introduce a log-log linearization method with shift correction (3LS) to estimate the exponent directly from the data without prior assumptions. This method employs a unique two-step optimization algorithm: it first assesses physical validity using linearity in the logarithmic space and subsequently evaluates statistical optimality using AIC in the original data space. This hybrid framework incorporates a hierarchical decision logic that verifies physical and statistical validity, automatically classifying result reliability. Verification using simulation data confirmed that this method identifies the true physical model more reliably than the conventional metric, and that its decision logic correctly flags the result as indistinguishable once noise makes the candidate models statistically inseparable. Application to standard gold samples yielded the expected n = 2 model with high reproducibility (Ith = 4.89 ± 0.01 eV). Finally, analysis of 27 diverse organic semiconductors demonstrates that this framework is a high-throughput tool for materials informatics.