In semiconductor manufacturing, anomaly detection of equipment and processes, and the cause analysis of product quality and measurement results are very important analytical activities. The observed data from the process and equipment can be understood as functions defined on the time domain. We propose a new method that classifies normal and abnormal products, exploiting the functional nature of the data. We develop a functional logistic regression method in which coefficient function is modeled by a constant spline to analyze wafer data in semiconductor etching processes. In order to detect change points in the process, we obtain a knot sequence used to construct a constant B-spline basis based on the second derivatives values of predictor functions. Based on the knot information, features were extracted from the inner product with a constant spline and applied to the generalized linear model. The abnormal wafers were classified based on the estimated coefficient function, and it was confirmed that the average variation of the segmented specific time interval of the predictor influences the wafer quality judgment. It is useful for interpreting the coefficient function on each time interval, then the proposed method is expected to be suitable for anomaly detection and cause analysis in the relevant fields.