Analysing the impact of Machine Learning Health Operations (MLHOps) on Mental Health and Stress

Venkatraj, Naaga Barani Govindan (2023) Analysing the impact of Machine Learning Health Operations (MLHOps) on Mental Health and Stress. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

This study examines how Machine Learning Health Operations (MLHOps) can be used to address the growing global mental health issues, which have been made worse by the COVID-19 pandemic. Acknowledging mental health as integral to overall well-being, the study emphasizes the urgent need for innovative solutions and early intervention. Leveraging the power of cloud computing, specifically Microsoft Azure, the research aims to utilize social media data, particularly Twitter, for early detection of mental health conditions, with a focus on depression. The dataset, collected from multiple Twitter accounts, was comprised of normal tweets and a labelled dataset of depressive tweets from users diagnosed with depression and other mental health conditions. Natural language processing (NLP) techniques were applied to extract features from the tweet text that could indicate signs of depression. The tweets were preprocessed by removing URLs, usernames, hashtags, and stopwords, then vectorized into dense word embeddings using the Spacy language model. Three supervised machine learning models were trained on the dataset to classify users as depressed or not depressed based on these text features - K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM). The system achieved an accuracy of 82% with RF, outperforming KNN at 78% and SVM at 85%. To operationalize the analysis, an end-to-end Azure Machine Learning (AML) pipeline was built comprising data ingestion, preprocessing, model training and evaluation to reflect the ML Health Operations (MLHOps) practices. This enables an automated workflow to continuously collect new Twitter data, extract features, and identify users at risk of depression. The methodology demonstrates the feasibility of monitoring population-level mental health from social media data. Future work includes expanding beyond depression to detect other conditions and improving model feasibility.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Makki, Ahmed UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Cloud computing P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning R Medicine > RA Public aspects of medicine > RA790 Mental Health
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	28 Mar 2025 09:47
Last Modified:	28 Mar 2025 09:47
URI:	https://norma.ncirl.ie/id/eprint/7345

Actions (login required)

View Item