Diabetes is one of the popular chronic diseases in the world. The long term diabetes increases the risk of the stroke, the heart diseases, the kidney damage and the blindness. Therefore, the blood glucose estimation is crucial for the diabetic patients. The photoplethesmograms (PPGs) can be used for performing the non-invasive blood glucose estimation. In particular, the features were extracted from the PPG represented in both the time domain and the frequency domain. Nevertheless, the total number of the features was very limited. Hence, the performance of the blood glucose estimation was poor. To address this issue, this paper proposes to employ the fractional Fourier transform for extracting the features in the rotated time frequency domain. Since more than one rotational angle can be performed and a set of features can be extracted from each rotational angle, the total number of the features is significantly increased. The computer numerical simulations are conducted. Compared to the method only using the time domain features and the frequency domain features, it is found that the conventional method without the fractional Fourier transform domain features yields 77.2% of all the test data in the Zone A of the grid, respectively. On the other group, our proposed method with the fractional Fourier transform domain features yields 88.6% of all the test data in the Zone A of the grid, respectively. This demonstrates the superiority of our proposed method.