Application of Machine Learning Algorithms to Emotion Recognition Using Physiological Information
| Autor: | LIN, LIANG-HAO, 林亮豪 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 The pressure due to heavy workload and life burden seriously affect people's physical and mental health. Monitoring people's emotion through emotional recognition is able to improve the quality of life, and even can save lives in certain situations. This research aims to develop a human emotion recognition system by use of the variations of the pulse rate, saturation of blood oxygen, and skin conductance. Moreover, those physiological information are transmitted to an Android smartphone via Bluetooth, and the physiological information are further transmitted to the remote server from the Android smartphone to build the database. The user’s emotion can be recognized by employing the proposed ERS with the dataset built in the server. Three kinds of motion that commonly happen in daily life are recognized, which are the happiness, angry, and sadness. Collecting 13400 physiological data from fifteen volunteers. Fourteen approaches are employed to extract the features for training and test. Four machine learning schemes, which consist of K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGboost) are applied to execute the emotion recognition task. The recognition performance is evaluated in terms of accuracy, precision, recall rate, specificity and F1-score. Experimental results show that the best performance can be achieved when all three physiological signals are employed together with a segment size of 45. In addition, the optimal parameters allocation for reducing the processing time is investigated. Experimental results of 10-fold cross validation show that the F1-score of Random Forest, XGboost, are respectively 99.81%, and 99.80%. Therefore, in terms of F1-score and processing time, Random Forest is suggested to be the top priority choice for emotion recognition. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|