Index normalization based probability distribution selection method for model selection

摘要

The present invention discloses a probability distribution selection method for improving the determination deviation of the conventional probabilistic method in which the selection is based on a conventional single model selection index in determining the representative probability distribution that best matches the measured values. A method for selecting a probability distribution based on normalization of a model selection index according to the present invention comprises: inputting relevant raw data of a selected medium to the control unit 10 (S10); selecting, by the operation unit 20, a plurality of candidate probability distributions including at least a main probability distribution function having a distribution similar to that of the raw data from among the probability distribution functions stored in the probability distribution DB 30 (S20); The calculation unit 20 uses at least one of a first model selection index having a test in which a relative index and an absolute index (p-value) is provided and a second model selection index without a test in which only a relative index is provided. calculating a model selection index (MSI) for evaluating the candidate probability distribution of (S30); calculating, by the calculator 20, a normalized relative index score (n-Score) by normalizing the relative index of the model selection index within a predetermined value range for each candidate probability distribution (S40); The calculating unit 20 calculates the final relative index score (F-Score), which is the sum of the relative index scores normalized for each candidate probability distribution (S50), and compares the final relative index scores to fit the candidate probability distribution of the highest final relative index score selecting a probability distribution (S60); and storing the appropriate probability distribution in the probability distribution DB 30 (S70).