izpis_h1_title_alt

Network representational learning for classification of patients with neurological disorders
ID NIKOLOSKA, BISERA (Author), ID Šubelj, Lovro (Mentor) More about this mentor... This link opens in a new window, ID Kastrin, Andrej (Comentor)

.pdfPDF - Presentation file, Download (1,43 MB)
MD5: 72CC715B25AE9D3E39BD7020DDC6DFE9

Abstract
Neurological disorders present challenges in the realms of diagnosis and treatment, primarily due to the multifaceted nature of symptoms exhibited by affected individuals. This thesis delves into the domain of network repre- sentational learning for patient classification within the context of neurolog- ical disorders. In general, we have a dataset with two groups of patients - the typically developing patients and the patients with a neurological disorder. We an- alyze these patients by constructing graphs in which each node symbolizes an individual patient within our dataset. In further analysis of these graph representations, we use network embedding techniques, where we derive low- dimensional node representations. We look into three different algorithms; node2vec, HOPE and matrices with the pertinent graphic characteristics. After experimenting with all the hyperparameters that the algorithms use, we choose node2vec for further analysis, since it delivers highest scores. Our study extends beyond the realm of graph embeddings by employing classification models. The embeddings serve as essential input features for machine learning algorithms, including Random Forests, Logistic Regression and Support Vector Machines (SVM). This ensemble of classifiers collectively facilitates the accurate classification of patients across various neurological disorders. From the three classification models, Random Forests yielded the best results, delivering an accuracy of 90% for node2vec. The research findings underscore the potential of network representation graphs as a tool for characterizing and classifying patients afflicted with neu- rological disorders. Our comparative analysis of classification models reveals the efficacy of graph-based embeddings in enhancing diagnostic accuracy.

Language:English
Keywords:complex networks, brain networks, graph embeddings, machine learning, classification
Work type:Bachelor thesis/paper
Typology:2.11 - Undergraduate Thesis
Organization:FRI - Faculty of Computer and Information Science
Year:2023
PID:20.500.12556/RUL-150261 This link opens in a new window
COBISS.SI-ID:168549379 This link opens in a new window
Publication date in RUL:15.09.2023
Views:444
Downloads:68
Metadata:XML DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:Slovenian
Title:Omrežno predstavitveno učenje za uvrščanje bolnikov z nevrološkimi motnjami
Abstract:
Diagnoza in zdravljenje nevroloških obolenj predstavlja poseben izziv, predvsem zaradi raznolike narave simptomatike. V diplomskem delu se ukvarjamo s problematiko uvrščanja nevroloških bolezni s pomočjo omrežnega predstavitvenega učenja. Obravnavamo nabor podatkov z dvema skupinama pacientov: skupino zdravih udeležencev in skupino bolnikov z diagnosticirano nevrološko motnjo. V nalogi izhajamo iz predpostavke, da lahko možgansko konektivnost predstavimo s pomočjo omrežja, v katerem se vozlišča nanašajo na izbrana področja možganske skorje, povezave med vozlišči pa predstavljajo moč konektivnosti. Jedro naloge temelji na vektorskih vložitvah vozlišč tako zgrajenih omrežij, nad katerimi smo zgradili klasifikatorje za uvrščanje posameznika bodisi v skupino pacientov z nevrološkimi obolenji bodisi v skupino zdravih udeležencev. Za vlaganje vozlišč smo preizkusili algoritma node2vec in HOPE. Za primerjavo smo v analizo vključili tudi uvrščanje na osnovi nekaterih atributov, s katerimi sicer opisujemo lastnosti kompleksnih omrežij. Za uvrščanje smo preizkusili tri metode strojnega učenja in sicer logistično regresijo, naključne gozdove in metodo podpornih vektorjev. Po eksperimentiranju s hiperparametri, ki jih uporabljajo ti algoritmi, smo za nadaljnjo analizo izbrali algoritem node2vec, ker je bil najučinkovitejši. V kombinaciji z naključnimi gozdovi smo pri uvrščanju dosegli 90\,\% točnost uvrščanja. Rezultati potrjujejo velik potencial predstavitvenega učenja nad možganskimi omrežji, kar smo potrdili z nalogo klasifikacijo pacientov z nevrološkimi obolenji. Opravljena primerjalna analiza klasifikacijskih modelov kaže na učinkovitost vključevanja omrežnih modelov možgan za izboljšanje diagnostične natančnosti.

Keywords:kompleksna omrežja, možganska omrežja, strojno učenje, uvrščanje

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Back