Earticle

This study is an extension of our last researches about the detection of transition zones based on multiresolution spectral analysis (MRS). In this paper we present the fourth step for the realization of an automatic system for Tunisian Dialect segmentation and analysis. The MRS is calculated over several Fast Fourier Transforms (FFT) of different length. It can provide a higher temporal accuracy in the upper spectral region and a better frequency resolution in the lower spectral range. We showcase the importance of this tool by attempt- ing to an automatic transition zones detection by calculating the Interquartile Range (IQR) of all frames of the MRS FFT. We applied our Visual Assistance of Speech Processing (VASP) System to a corpus. This corpus was in Tunisian Dialect pronounced by tunisian speakers. The analysis of the obtained results shows that the automatic detection of tran- sition zones based on MRS provides better results compared to classical spectral analysis of the corpora used.

This paper presents an automatic method for detecting transition zones based on multires- olution spectral analysis (MRS). The MRS is calculated over several Fast Fourier Trans- forms (FFT) of different length. It can provide a higher temporal accuracy in the upper spectral region and a better frequency resolution in the lower spectral range. We showcase the importance of this tool by attempting an automatic detection of zones of transition by calculating the Interquartile Range (IQR) of each frame of the MRS FFT. We applied our Visual Assistance of Speech Processing (VASP) System to a corpus. This corpus was in French pronounced by french speakers and has the format CiV CiV with Ci was a stop con- sonant [p t k] and V was a vowel [i e]. The results showed that the automatic detection of transition zones based on MRS provides better results compared to classical spectral analysis of the corpora used.

최근 전 세계적으로 당뇨병 유발률이 증가함에 따라 다양한 머신러닝과 딥러닝 기술을 통해 당뇨병을 예측하려고 는 연구가 이어지고 있다. 본 연구에서는 독일의 Frankfurt Hospital 데이터로 머신러닝 기법을 활용하여 당뇨병을 예측하는 모델을 제시한다. IQR(Interquartile Range) 기법을 이용한 이상치 처리와 피어슨 상관관계 분석을 적용하고 Decision Tree, Random Forest, Knn, SVM, 앙상블 기법인 XGBoost, Voting, Stacking로 모델별 당뇨병 예측 성능을 비교한다. 연구를 진행한 결과 Stacking ensemble 기법의 정확도가 98.75%로 가장 뛰어난 성능을 보였다. 따라서 해당 모델을 이용하여 현대 사회에 만연한 당뇨병을 정확히 예측하고 예방할 수 있다는 점에서 본 연구는 의의가 있다.

With the recent increase in diabetes incidence worldwide, research has been conducted to predict diabetes through various machine learning and deep learning technologies. In this work, we present a model for predicting diabetes using machine learning techniques with German Frankfurt Hospital data. We apply outlier handling using Interquartile Range (IQR) techniques and Pearson correlation and compare model-specific diabetes prediction performance with Decision Tree, Random Forest, Knn (k-nearest neighbor), SVM (support vector machine), Bayesian Network, ensemble techniques XGBoost, Voting, and Stacking. As a result of the study, the XGBoost technique showed the best performance with 97% accuracy on top of the various scenarios. Therefore, this study is meaningful in that the model can be used to accurately predict and prevent diabetes prevalent in modern society.

표본크기를 결정할 때 모표준편차 $\sigma$의 추정량으로 표본표준편차를 구할 수 없는 경우 범위(R)또는 사분위간 범위(IQR)를 이용하여 $\sigma$의 추정량으로 사용할 수 있다 R과 IQR의 함수로 나타난 추정값은 최소한 95% 이상의 확률로 $\sigma$보다 크거나 같아야 과소 추정됨을 피할 수 있다. 다양한 확률분포로부터 추출된 여러 표본의 범위와 사분위간 범위에 대하여 Browne(2001)이 연구한 추정량 R/4과 본 연구에서 제시한 추정량 IQR이 $\sigma$이상일 확률에 대하여 비교 분석을 하였다. 그리고 표본의 범위와 사분위간 범위를 상수로 나누었을 때 $\sigma$이상일 확률을 가질 수 있는 대안적 인 분모를 각각 구하여 비교 연구하였다.

Without a sample standard deviation for an estimator of the population standard deviation u in a sample size computations, we often use some functions of a sample range (R) or interquartile range (IQR) by an estimator of $\sigma$. In order to avoid under-powered studies, these estimates must have a high probability of being greater than or equal to $\sigma$. In this paper, these probabilities of being greater than or equal to $\sigma$ are estimated for IQR for various parents distributions, and are compared with the probabilities for R/4 (Browne 2001). Alternative divisors (K) are explored and discussed for which the probabilities of R/K and IQR/K being greater than or equal to $\sigma$ is at least 95%.