KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Settings

Has files: NoSubject: search.filters.subject.imbalanced data

Search Results

Now showing 1 - 3 of 3
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Effect of Class Weights on Imbalanced Classes in Bi-directional LSTM Training for Sleep Apnea Classification
    (2024-01-01)
    Jansri, Ukkrit
    ;
    Tretriluxana, Suradej
    Sleep apnea, which is defined as the repetitive cessations of breathing during sleep, is the common disorder worldwide. The cost and the process of sleep test to obtain the polysomnogram is not optimal for sleep apnea screening in the large population. A deep learning model was developed to classify the normal and apnea events in a single time-series signal exported from the US National Institute of Health (NIH) sponsored database. Our challenge was to train the model with imbalanced dataset between normal and abnormal respiratory events. Three different methods, Synthetic Minority Over-sampling Technique (SMOTE), Random Under-Sampling (RUS), and the Class Weights (CW) were chosen to improve the model performance over the original data on five selected signals from polysomnographic dataset. The binary classification outputs were evaluated by four metrics. Our results showed (1) Matthews Correlation Coefficient was highest (MCC= 0.1385) in the Class Weights method on the nasal airflow signal. (2) Cohen's Kappa score, was highest (k= 0.0819) in SMOTE technique on the abdominal signal, followed by the Class Weights method on the abdominal signal (k= 0.0687) and RUS technique on nasal airflow signal (k= 0.0441). (3) F1-score was highest (F1= 11.89%) in SMOTE technique on the abdominal signal, followed by the Class Weights method on nasal airflow signal (F1 = 11.17%) and RUS technique on nasal airflow signal (F1= 9.16%). The findings suggest that the Class Weights method on nasal airflow and the Class Weights method on abdominal signal were the two combinations to be used in the DL model.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Application of Natural Neighbor-based Algorithm on Oversampling SMOTE Algorithms
    (2021-04-01)
    Srinilta, Chutimet
    ;
    Kanharattanachai, Sivakorn
    Classification performance depends highly on data distribution. In real life, data often come imbalanced where one class is found more often than others. SMOTE-based algorithms are usually used to handle the class imbalance problem. One key parameter that algorithms in SMOTE family require is k-the number of nearest neighbors with respect to a certain data point. K that fits the dataset the most gives the optimum performance. This paper proposes an approach to suggest a value of the parameter k using Natural Neighbor algorithm. Datasets are made balanced by four SMOTE-based algorithms-standard SMOTE, Safe-Level-SMOTE, ModifiedSMOTE and Weighted-SMOTE. The F-measure and Recall matrices are used to evaluate classification performance of a Support Vector Machine classifier running against six datasets with different imbalance ratios. The results show that, the average classification performance achieved by the proposed k's is closer to the optimum when compared with the performance given by the default value of k.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Fast Learning and Testing for Imbalanced Multi-Class Changes in Streaming Data by Dynamic Multi-Stratum Network
    (2017-01-01)
    Thakong, Mongkhon
    ;
    Phimoltares, Suphakant
    ;
    Jaiyen, Saichon
    ;
    Lursinsap, Chidchanok
    Although several efficient learning methods have recently been proposed to handle class drift situations, issues remain in various streaming data applications that possibly deteriorate classification accuracy. Three important issues were considered, that is: 1) lifetime and class changes; 2) high imbalance ratios of streaming data among classes; and 3) classification accuracy of untrained data and class-changed data. A new dynamical learning structure based on hyper-elliptical capsule and multi-stratum network was introduced to cope with these issues. The experimental results on a simulated University of California at Irvine non-concept-drift database and real concept-drift data confirm that the proposed multi-stratum learning provided better accuracy, faster learning speed, and lower structural complexity than other concept-drift algorithms.