In the last few years, the clustering of time series has seen significant growth and has proven effective in providing useful information in various domains of use. This growing interest in time series clustering is the result of the effort made by the scientific community in the context of time data mining. For these reasons, the first phase of the thesis focused on the study of the data obtained from fMRI exams carried out in task-based and resting state mode, using and comparing different clustering algorithms: SelfOrganizing map (SOM), the Growing Neural Gas (GNG) and Neural Gas (NG) which are crisp-type algorithms, a fuzzy algorithm, the Fuzzy C algorithm, was also used (FCM). The evaluation of the results obtained by using clustering algorithms was carried out using the Davies Bouldin evaluation index (DBI or DB index). Clustering evaluation is the second topic of this thesis. To evaluate the validity of the clustering, there are specific techniques, but none of these is already consolidated for the study of fMRI exams. Furthermore, the evaluation of evaluation techniques is still an open research field. Eight clustering validation indexes (CVIs) applied to fMRI data clustering will be analysed. The validation indices that have been used are Pakhira Bandyopadhyay Maulik Index (crisp and fuzzy), Fukuyama Sugeno Index, Rezaee Lelieveldt Reider Index, Wang Sun Jiang Index, Xie Beni Index, Davies Bouldin Index, Soft Davies Bouldin Index. Furthermore, an evaluation of the evaluation indices will be carried out, which will take into account the sub-optimal performance obtained by the indices, through the introduction of new metrics. Finally, a new methodology for the evaluation of CVIs will be introduced, which will use an ANFIS model.

Analysis of FMRI Exams Through Unsupervised Learning and Evaluation Index / Samuele Martinelli , 2020. 33. ciclo, Anno Accademico 2019/2020.

Analysis of FMRI Exams Through Unsupervised Learning and Evaluation Index

MARTINELLI SAMUELE
2020-01-01

Abstract

In the last few years, the clustering of time series has seen significant growth and has proven effective in providing useful information in various domains of use. This growing interest in time series clustering is the result of the effort made by the scientific community in the context of time data mining. For these reasons, the first phase of the thesis focused on the study of the data obtained from fMRI exams carried out in task-based and resting state mode, using and comparing different clustering algorithms: SelfOrganizing map (SOM), the Growing Neural Gas (GNG) and Neural Gas (NG) which are crisp-type algorithms, a fuzzy algorithm, the Fuzzy C algorithm, was also used (FCM). The evaluation of the results obtained by using clustering algorithms was carried out using the Davies Bouldin evaluation index (DBI or DB index). Clustering evaluation is the second topic of this thesis. To evaluate the validity of the clustering, there are specific techniques, but none of these is already consolidated for the study of fMRI exams. Furthermore, the evaluation of evaluation techniques is still an open research field. Eight clustering validation indexes (CVIs) applied to fMRI data clustering will be analysed. The validation indices that have been used are Pakhira Bandyopadhyay Maulik Index (crisp and fuzzy), Fukuyama Sugeno Index, Rezaee Lelieveldt Reider Index, Wang Sun Jiang Index, Xie Beni Index, Davies Bouldin Index, Soft Davies Bouldin Index. Furthermore, an evaluation of the evaluation indices will be carried out, which will take into account the sub-optimal performance obtained by the indices, through the introduction of new metrics. Finally, a new methodology for the evaluation of CVIs will be introduced, which will use an ANFIS model.
2020
UNSUPERVISED LEARNING, CLUSTERING , fMRI, EVALUATION INDEX CVIs, MACHINE LEARNING,
Analysis of FMRI Exams Through Unsupervised Learning and Evaluation Index / Samuele Martinelli , 2020. 33. ciclo, Anno Accademico 2019/2020.
File in questo prodotto:
File Dimensione Formato  
Martinelli_Tesi_PHD_03_12_2020_pdfa.pdf

accesso aperto

Tipologia: Tesi di dottorato
Licenza: Dominio pubblico
Dimensione 6.96 MB
Formato Adobe PDF
6.96 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11383/2115066
 Attenzione

L'Ateneo sottopone a validazione solo i file PDF allegati

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact