With the increase in the instances of damage caused by counterfeits of electronic components, a Physical Unclonable Function (PUF) has attracted attention as a technique to prevent counterfeiting. PUF uses arbitrary dispersion as specific identification, which is generated during semiconductor manufacturing. Even if all LSI circuit patterns are copied, PUF can discriminate the LSI circuit from counterfeits. However, the vulnerability of PUF to machine learning attacks has been reported. Although machine learning attacks can predict responses to unknown challenges with high probability, these attacks cannot completely predict responses. For machine learning attacks, the present study proposed a method to clone PUF's responses using error correction techniques, such as Fuzzy Extractor. The present study also demonstrated that the proposed method could clone PUF mathematically. An evaluation experiment revealed that in machine learning attacks, when the number of learning times was 1,024, the ratio of an arbiter PUF with 64 selector steps to predict responses was 95% and that of the proposed method was 100%. Therefore, the proposed method could clone responses completely..