[dipl] Perustieteiden korkeakoulu / SCI

Permanent URI for this collectionhttps://aaltodoc.aalto.fi/handle/123456789/21

Browse

Recent Submissions

Now showing 1 - 20 of 5242
  • Large Language Model Agent as Insurance Law Assistant
    (2024-09-30) Ingher, Amir
     School of Science | Master's thesis
    Traffic insurance law can be a complex domain for ordinary individuals to grasp, thus affecting its accessibility. Although the traffic insurance law itself and precedents are publicly available, it can be challenging to navigate through this information in order to understand one’s rights for compensation in the event of a traffic accident. Recent advancements in Large Language Models (LLM) have provided an accessible way for people to ask curated questions, that the LLM then answers. Thus, the thesis aims to utilize LLM and surrounding technologies to improve the accessibility of traffic insurance law for everyday users trying to navigate this complex legal domain. The research was conducted as a project for an early-stage startup. The thesis developed an agent that employs Retrieval-Augmented Generation (RAG) techniques with LLM. RAG retrieves relevant information from customselected legal documents and uses the information to answer users’ questions. The agent is encapsulated within a web application which provides an accessible way for the user to communicate with the agent. The results were evaluated by a legal expert in a human-feedback manner. The expert indicated that the system could generate satisfactory responses to experts’ questions. By employing RAG techniques the system mitigated hallucination which is commonly related to LLMs, therefore improving the quality of the answer. However, the research also found that the document retrieval was suboptimal, thereby needing more work to improve the retrieval process. The study highlights the potential of using LLMs and RAG techniques to make traffic insurance law more accessible to common people. Furthermore, the research suggests that improving the retrieval mechanism and possibly employing a multi-agent system could further improve the contextual understanding of the system and thus aid individuals in navigating legal processes more effectively. Hence, future research will focus on optimizing retrieval processes and exploring the use of multiple agents.
  • Digitalization’s effect on port business
    (2024-09-26) Pallasaho, Tommi
     School of Science | Master's thesis
    This thesis investigates the impact of digitalization on the port industry, focusing on how digital technologies are reshaping port operations, business models, and future opportunities. A qualitative research methodology was employed, utilizing semi-structured interviews with digitalization experts, sales personnel, and top management from a leading company in the industry. The data collected were analysed using thematic analysis. The findings reveal that while the port industry has begun embracing advanced technologies, it significantly lags behind sectors like aviation and rail. The industry's conservative nature, fragmentation, and resistance to change are major barriers to digital transformation. Key challenges identified include cultural shifts, workforce adaptation, data quality issues, interoperability problems, and cybersecurity concerns. Technological advancements are poised to transform port operations by enhancing efficiency, predictability, and sustainability. The practical implications suggest that ports should develop comprehensive digital strategies, invest in workforce training, improve collaboration and data sharing, prioritize cybersecurity, and focus on sustainability. For Original Equipment Manufacturers (OEMs), the study highlights the need for innovation in product development, offering integrated solutions, adopting a customer-centric approach, and forming strategic partnerships. Digitalization presents significant opportunities for the port industry to enhance operational efficiency, competitiveness, and sustainability. By addressing the identified challenges and strategically embracing technological advancements, both ports and OEMs can capitalize on the benefits of digital transformation, contributing to a more resilient and advanced maritime secto
  • Improving Single-Tenant SaaS Release Deployment Validation: a Case Study
    (2024-09-18) Tissari, Ville
     School of Science | Master's thesis
    Upgrading a single-tenant SaaS application to a newer version is a non-trivial task that requires time and resources from both the customer and the service provider. The upgrade process can be quite lengthy with multiple validation steps and people from different teams involved. In this thesis, we study a Finnish SaaS company that provides single-tenant software for their customers. The company has a centralized team performing the updates and the internal validation of the updates. The process is relatively time-consuming and involves repetitive manual steps and multiple data sources. The goal of this research was to identify the potential points of improvement in the process, construct a research artifact that would address one or more of these points, and test the artifact in practice. The research was conducted by utilizing the design science research paradigm. A round of semi-structured interviews was conducted in order to discover the potential areas of improvement for the existing upgrade process. Based on the interview results, it was hypothesized that reducing the amount of manual work involved in the validation step of the upgrade would improve the quality of the process. A Grafana-based tool was implemented as a design artifact and piloted in three separate upgrades to test this hypothesis. After the pilots, feedback was gathered from the engineers utilizing the tool, and a workshop was conducted to discuss the results. A literature survey was conducted in order to tie this research into existing research and to compare the problems encountered by the case company to the issues that are associated with single-tenant SaaS version upgrades specifically. The results of the pilot suggest that reducing the amount of manual work improves the quality of the upgrade process. Multiple improvements for the artifact were also proposed and left for future development. The improvement points discovered in the case company were compared to a multitude of sources that examine the attributes of single- and multi-tenant SaaS software. Challenges specific to single-tenant SaaS version upgrades were identified this way. The research suggests that customization is one of the main causes of issues in the upgrade process, leading to complexity in test suite maintenance, increased skill requirements for both the service provider and the customer, and increased risk of incidents leading to a decrease in service quality.
  • Simulations of quantum transport with quadratic fermionic Hamiltonians
    (2024-09-13) Mäkinen, Ilari
     School of Science | Master's thesis
    In this Master’s thesis, a simulation method for studying quantum transport with quadratic fermionic Hamiltonians is implemented and it is applied for studying the problem of heat transport through a single fermionic mode. The discussion begins with a review of quantum statistical physics and heat transport. Developing on this background, the single fermionic level heat transport problem is introduced and the basic properties of quadratic systems are reviewed. The theoretical basis of the simulation method based on general Bogoliubov transformations is then covered and the method is implemented for the single fermionic level heat transport problem. The validity of the implemented simulation is successfully evaluated by comparing the results to known expressions for steady state heat transport both in the weak coupling regime governed by a Lindblad master equation and the general coupling regime governed by a Landauer type formula. Furthermore, the effects of the density of states and the rotating wave approximation are discussed.
  • Controlling the Text Generation of a Large Language Model in Multilingual Setting using Latent Space Steering
    (2024-09-30) Leino, Julius
     School of Science | Master's thesis
    Numerous real-world applications of large language models (LLMs) require controlling the generated text to align with specific attributes. Furthermore, as LLMs are increasingly deployed in multilingual contexts, some completely new attributes emerge that must be controlled, along with novel challenges in controlling the same attributes as in English-specific settings. One of the new attributes that emerge is ensuring that the generated text is in a specific language. In this thesis, we show that a family of techniques referred to as latent space steering methods can be used to control the language of the text generated by an LLM, using Finnish and English as the case study. More specifically, we construct steering vectors from the intermediate representations of translation pairs and use these steering vectors to generate Finnish answers to challenging English questions, managing to shift the language in all responses to Finnish with a small degradation in the quality of the answers. In addition to ensuring that the generated text is in a specific language, another unique challenge to multilingual settings emerges from constructing the steering vectors for other attributes due to the creation requiring examples of the desired behaviour, which can be difficult to obtain with less represented languages. We provide one solution to this by showing that the effectiveness of some style steering vectors generalizes across languages. To demonstrate this, we use steering vectors for two styles constructed from English prompts to control the style of Finnish text generation. We observe that while there is a drop in the effectiveness of the steering vectors for the other style, the steering vectors for the other style achieve a near-perfect generalization performance. By further exploring these latent space steering methods in multilingual settings, we contribute to the important path of making the advancements in LLMs more accessible for the entire world.
  • Optimised Power Purchase Agreements Between Green Hydrogen & Green Electricity Producers
    (2024-09-30) Vuorinen, Mikael
     School of Science | Master's thesis
    The EU has decided to set a climate goal to be carbon-neutral before 2050, and the goal of Finland is to achieve this already by 2035. To meet these targets, the energy production sector must do huge reforms. Green hydrogen is promising technology which can be used to meet the targets. Green Hydrogen is produced using electricity from renewable energy sources. Power Purchase Agreement (PPA) is needed to meet the supply and demand of green electricity between these two parties. This research has three main goals: 1) to optimize PPA (Power Purchase Agreement) models which are designed for green electricity and green hydrogen producers. Focusing on the Finnish electricity market, 2) to understand the effect of flexibility options in the green hydrogen value chain on forming PPA models, and 3) to identify the influence of EU legislation on the content of PPA models. This research includes both qualitative and quantitative methods. The research consists of three different phases: 1) forming PPA models and evaluating the impact of hydrogen value chain flexibility possibilities and EU legislation on PPA models. This was done by doing a systematic literature review and semi-structured interviews. 2) After that the second round of interviews was done to evaluate the reliability of the preliminary results. Based on interviews five PPA models were selected for further studies. 3) Profitability calculations were done to the selected five PPA models using a data-driven analysis. Finally, conclusion was done based on results. The results revealed that hydrogen producers have enough flexibility possibilities to adapt to the fluctuating electricity markets. Based on the data analysis, the optimal PPA model was the "First Megawatt" structure with dynamic price flexibility. This proved to be the best option for green hydrogen producers. This research gives understanding of PPA structures between green electricity producers and green hydrogen producers in Finland and other countries. Additionally, this research provides valuable knowledge of the flexibility possibilities in the hydrogen value chain and gives knowledge of the impact of EU legislation on PPA models. The most significant achievement of this research is the creation of the potential PPA models and to give the recommendation of a PPA model between green electricity and green hydrogen producers.
  • Structural determinants of Griffiths phase in human brain dynamics
    (2024-09-09) Sormunen, Aleksi
     School of Science | Master's thesis
  • Planning a Solution for Mammography Data Management: Defining User Needs and Requirements
    (2024-09-29) Rytkönen, Helka
     School of Science | Master's thesis
    The future of medical technology is going towards data driven approaches and in the centre of it there is the management of large volumes of data. Managing DICOM files requires an efficient storage and the system should support efficient queries of the images. Currently, PACS is the most popular system in healthcare organizations. However, PACS does not always meet the requirements of all users in medical field such as companies that develop and product medical devices. In the literature, there are open-source tools for medical data management. Additionally, there exist companies that provide custom solutions for different needs. In cases where PACS is not a viable solution, these alternatives can be further explored. This study is a case study and the goal was to create a plan for an efficient internal data management solution for mammography images for a medical technology company. The study was conducted as a qualitative study. Data was collected from interviews with professionals with experience in medical data and its management. In addition, the company's needs were specified and used as a guide when writing requirements for the data management solution. All the interviewees had different perspectives and different solution tackling the challenges in data management. The important features that was noted from the interviews included accessibility, flexibility and privacy. With the help of requirements defined it was determined that the most optimal plan is to conduct a market research and find a company providing a solution that can be adapted. Finally, after conducting a focused market research a set of possible providers were discovered.
  • Prediction of Postoperative Pain with Intraoperative Photoplethysmogram
    (2024-09-14) Robertson, Emilia
     School of Science | Master's thesis
    Optimal analgesic administration during surgery is crucial to prevent both pain and excessive medication use. Predicting postoperative pain during surgery could enhance real-time drug administration, leading to better management and mitigation of postoperative pain. This thesis presents the development of a machine learning model for predicting postoperative pain, classifying patients into two categories: no pain and moderate-to-severe pain. Physiological data recorded during surgery were used to derive input features, while output labels were based on a self-evaluated pain score ranging from 0 to 10. The focus was on extreme pain scores (0 and 6--10) to ensure clear classification. Features were extracted from the Surgical Pleth Index (SPI) signal and its components, photoplethysmographic pulse wave amplitude (PPGA) and pulse rate (PR). The final input features included various statistical moments derived from the PPGA and PR signals. The training dataset comprised 78 patients, while the primary test set included 29 patients with extreme pain scores. Logistic regression was identified as the most suitable model. An aggregate measure calculated from model predictions throughout the surgery proved to be the most effective for assessing postoperative pain, achieving a PR AUC of 0.73 and an ROC AUC of 0.71 on the test set that included only extreme pain scores. Stable performance of the measure was observed during the last 40 minutes of surgeries. The results demonstrate the potential for predicting postoperative pain during surgery. This thesis introduced novel features and a method for continuous monitoring of postoperative pain risk. Future improvements include increasing the patient sample size and optimising the model's classification threshold.
  • Improving the compliance process for a medical device software using automation
    (2024-09-29) Hartikainen, Henni
     School of Science | Master's thesis
    The field of medical devices is highly regulated, with legislation and standards that must be followed to ensure the safety and effectiveness of the products. For a company developing a medical device, ensuring it follows the legislation and industry standards of the markets it operates in is critical. This compliance process includes verifying that safety-related testing has been correctly executed and documented. If done manually, this process can be time-consuming and prone to errors. This thesis investigates whether automation can be employed to improve this process of ensuring standards and regulations adherence for a software medical device. The objective of this thesis is to develop an automation tool that can be used to ensure the traceability and testing integrity of the safety-related design inputs of the medical device. In addition to reducing effort and avoiding the risk of manual errors, such an automation tool would enable the continuous monitoring of the traceability and testing integrity throughout the project. The development of the tool followed a product development process, beginning with defining the intended use and requirements for the tool, followed by implementation and finally validation of the tool. The tool was developed using Python. It extracts the necessary information from documents provided by the user and dynamically from a Test Management System via a REST API. The tool performs the necessary checks on the data and presents the results and any findings to the user. The performance of the tool was evaluated against the predefined requirements and by using it for its intended purpose during an actual product development project. The tool fulfilled all the requirements. When used for a product development project, it uncovered several findings from the data, including some unexpected inconsistencies. The tool clearly improves the efficiency of these checks: the time it takes from one person to perform the checks manually is in the order of days to weeks, whereas the tool takes less than 30 minutes to run. This improved efficiency allows the continuous monitoring of the verification status throughout the project. The tool proved itself useful in also other verification documentation tasks. These results together with the highly positive user feedback strongly suggest that automation can improve the efficiency of ensuring compliance with standards and regulations.
  • Construction of Hyper-Relational Knowledge Graphs Using Pre-Trained Large Language Models
    (2024-09-12) Datta, Preetha
     School of Science | Master's thesis
    Hyper-relational knowledge graphs serve as a technique to organize previously unstructured data. Question-answering systems built on these graphs excel at handling multi-hop questions and offer clear, transparent answers. However, developing a question-answering system centered around knowledge graphs can pose significant challenges and demands considerable effort. This thesis endeavors to streamline the process by leveraging large language models to generate hyper-relational knowledge graphs since it implies cheaper knowledge graph construction methodologies in the future. This thesis tests a range of prompting strategies across a subset of large language models to thoroughly evaluate their effectiveness in extracting entities and relations. These entities and relations are essential building blocks for constructing a knowledge graph. By applying different prompting techniques, the research aims to determine the most efficient and accurate methods for entity and relation extraction. This evaluation provides insights into the capabilities and limitations of large language models in the context of knowledge graph development. We also perform a comparison of the prompting techniques with some existing supervised methodologies. The evaluation metric utilized in this thesis is BERTScore. Additionally, the thesis provides a comprehensive discussion on the advantages and disadvantages of BERTScore, as well as other evaluation metrics. This analysis aims to highlight the strengths and limitations of each metric, offering a balanced perspective on their applicability and effectiveness in assessing the outcomes of entity and relation extraction. The highest results achieved in this thesis are attributed to large language model based prompting that incorporates the relation descriptions of the dataset.
  • An Evaluation Framework for Assessing Uses of Radio Telescopes
    (2024-09-30) Emelianov, Oskar
     School of Science | Master's thesis
    This thesis presents an evaluation framework for assessing the uses and impacts of radio telescopes, leveraging Multi-Attribute Value Theory (MAVT). The framework is designed to evaluate the scientific, technological, educational and societal contributions of radio telescopes, providing a structured approach for decision-making. The case study of the Atacama Large Millimeter/submillimeter Array (ALMA) demonstrates the framework’s applicability and effectiveness. ALMA’s high value score illustrates its potential to advance astronomical research, foster international collaboration and enhance educational outreach. While the framework demonstrates its adaptability, it serves as a structured tool for decision-making rather than offering definitive conclusions about broader impacts. The framework’s adaptability to different contexts and time frames ensures its relevance in the dynamic field of radio astronomy, promoting balanced and informed decisions that consider both short-term gains and long-term strategic goals. This thesis underscores the importance of comprehensive assessment tools in managing large-scale scientific infrastructures, paving the way for future research and technological advancements.
  • Algorithms for Network Design
    (2024-09-30) Stimpert, Ronja
     School of Science | Master's thesis
    The increasing demand for storing enormous amounts of data while accessing it quickly caused the development of datacenters with flexible connections, so called demand-aware networks. In order to facilitate such networks optimally research is conducted to find algorithms embedding a network in another, sparser, network, or to conclude certain algorithms cannot exist. This thesis is investigating the growth of distances between neighboring nodes when embedding networks: we take a network G and state certain requirements to a host network G∗, to store all information of G while changing the connections. When rearranging the connections we want to minimize the distances created between network nodes that were formerly connected. In the presented work we first show this for transitions between the graph families cycles, paths, and trees, leading up to finding a network with degree ∆ that cannot be embedded in any tree without increasing the distances of at least some neighbors to an order of magnitude of ∆. The results provide on the one hand some basic methods, lower and upper bounds useful for researching transitions not investigated here, on the other hand we found a lower bound of Omega(∆) for the created local distances when hosting general graphs with n nodes in trees and an outline for improving this result to Omega(log n).
  • A biobjective mixed-integer linear program for decarbonized bunkering locations in the Baltic sea
    (2024-09-26) Österbacka, Mattias
     School of Science | Master's thesis
    Maritime shipping accounts for almost 3% of yearly global greenhouse gas emissions. One potential way to reduce these emissions is to switch from burning oil to renewable synthetic fuels. Hydrogen and ammonia can be produced in a renewable way, and can be fed into a fuel cell to directly generate electricity to power a ship. Ammonia can also be burned in engines like traditional marine fuels, aiding the potential switch. In this thesis, a biobjective mixed-integer linear program is developed and implemented to aid in optimally placing renewable bunkering locations in the Baltic Sea. A price model for producing hydrogen and ammonia around the Baltic Sea, as well as historical ship data is used in conjunction with the optimization model. Several solution approaches for multiobjective optimization problems are implemented and used to compute Pareto optimal solutions. A robustness measure for production locations, called the core index, is applied. Based on the robustness measure, we identify ports where producing renewable bunkering fuel is advisable, even if the preference between reducing emissions and keeping costs down is not known in advance. Sensitivity analysis is performed to gauge how the results are impacted by changes in input parameters.
  • Improving inventory level control of spare parts: Case study in machinery manufacturing
    (2024-09-30) Välisalmi, Mikko
     School of Science | Master's thesis
    Managing spare parts inventories is challenging due to intermittent nature of demand and ever-increasing customer expectations for service responsiveness. The case company operating in the machinery manufacturing industry has undergone rapid growth in aftersales and is now seeking to optimize their inventory levels by managing the trade-off between service level and working capital committed to inventories. Optimizing inventory levels requires setting inventory management practices, conducting demand forecasting, and establishing service level targets to arrive at economically appropriate safety stock levels for spare parts. The objective of this thesis was to develop a methodology for improving inventory level planning and control process in the context of intermittent demand of spare parts. In this study, first, a framework for developing understanding into a company’s current inventory level control practices is provided and an approach for solving complications in the process is presented. Then, the methodology is applied in the case context to offer evidence for the method. The study was conducted in two parts. First, a literature review was conducted to research relevant frameworks and concepts around the topic. Then, an empirical case study was executed to establish the state of current inventory level control practices and complications, and to solve for the discovered complications. The results of this study first highlight a three-phase framework for improving inventory level control process. Then the framework is used to show that there are six inventory level control practices that the case company is currently doing in accordance to approaches suggested in literature. There are also four practices that are seen as complications at the case company. This thesis offers solutions to all discovered complications along with options for further improvement. The methodology and the results in this thesis can be applied in other similar contexts with intermittent demand patterns of spare parts.
  • Enhancing Telecom Software Development Processes: A Framework for Customer-Centric Requirements Elicitation and Validation Processes
    (2024-09-29) Mohamud, Mubashir
     School of Science | Master's thesis
    This thesis focuses on enhancement of software development processes in the content of telecommunications industry through proposal of a customer-centric requirements elicitation and validation framework. The telecom sector along with its unique challenges with stringent reliability, security, and regulatory compliance, demands a specialized approach to software development. This thesis identifies key challenges in eliciting and validating customer requirements in the telecom software development processes. Through the case study of a telecom company, this thesis identified key challenges in customer-centric requirements engineering. These challenges include limited resources and the absence of standardized documentation, which hinder the effectiveness of requirements elicitation and validation. To address these challenges identified from interview findings, the thesis suggests a framework that integrates techniques such as user stories to capture detailed customer requirements while maintaining the broader context and rationale. It emphasizes the use of standardized documentation templates to ensure comprehensive and consistent recording of requirements. For validation, walkthroughs and mockups are used to assess feature feasibility and verify understanding from the customer's perspective. Additionally, a traceability matrix is maintained to keep track of changes systematically, ensuring efficient resource management even in the case of limited time and budget constraints. The proposed framework helps telecom companies better align their software development processes with customer needs, improving their satisfaction and reducing the risk of software misalignment. It offers a structured, flexible, and repeatable approach to requirements engineering. This can significantly enhance both customer value and the efficiency of the development lifecycle. The thesis concludes with recommendations for future research to further refine and test the framework across different telecom organizations.
  • Creating product portfolio strategy in multinational companies
    (2024-09-30) Koljonen, Markus
     School of Science | Master's thesis
    Multinational companies face the complex challenge of managing diverse product portfolios across various markets. Effective Product Portfolio Strategy (PPS) is essential for aligning products with corporate objectives, optimising resources, and maintaining competitiveness. However, there is limited research on how these companies create PPS and the organisational factors that influence it. This thesis explores how multinational companies develop their product portfolio strategies and identifies key organisational factors affecting them. Guided by two research questions – (1) How do companies create product portfolio strategy? and (2) What organisational factors affect product portfolio strategy? – the study adopts an inductive qualitative approach. Semi-structured inter-views were conducted with 18 professionals from nine multinational companies. Data was analysed using the Gioia methodology. Findings reveal that companies employ a structured, top-down approach to create PPS, beginning with comprehensive evaluations of external and internal environments. Small, cross-functional teams facilitate efficient decision-making and ensure alignment. The resulting strategies include detailed roadmaps and clear business objectives. Key organisational factors influencing PPS include organisational structure, interdependencies, alignment within the portfolio, and existing portfolio management practices. Aligning organisational structures with the portfolio and effectively managing interdependencies are crucial for successful strategy development. Utilising global products over local variants enhances efficiency and optimises resource allocation. This study fills a gap in the literature on PPS in multinational firms and offers practical insights for managers. By adopting structured approaches and aligning organisational factors, companies can enhance their strategic capabilities and achieve competitive advantage in the global market.
  • Win probability estimation for strategic decision-making in esports
    (2024-08-26) Jalovaara, Perttu
     School of Science | Master's thesis
    Esports, i.e., the competitive practice of video games, has grown significantly during the past decade, giving rise to esports analytics, a subfield of sports analytics. Due to the digital nature of esports, esports analytics benefits from easier data collection compared to its physical predecessor. However, strategy optimization, one of the focal points of sports analytics, remains relatively unexplored in esports. In traditional sports analytics, win probability estimation has been used for decades to evaluate players and support strategic decision-making. This thesis explores the use of win probability estimation in esports, focusing specifically on League of Legends (LoL), one of the most popular esports games in the world. The objective of this thesis is to formalize win probability added, i.e., the change in win probability associated with a certain action, as a contextualized measure of value for strategic decision-making, using mathematical notation appropriate for contemporary esports. The proposed method is elaborated by applying it to the evaluation of items, a strategic problem in LoL. To this end, we train a deep neural network to estimate the win probability at any given LoL game state. This in-game win probability model is then benchmarked against similar models.
  • Data mining and building algorithms for disease severity prediction in multiple sclerosis
    (2024-09-29) Oksanen, Olli
     School of Science | Master's thesis
    Multiple sclerosis (MS) is a neuroinflammatory and demyelinating disease targeting the central nervous system (CNS). The disease affects millions of people worldwide. The symptoms of MS are heterogeneous and often caused by the demyelinating lesions in the CNS. Thus, understanding factors, which can predict or cause disease progression and worsening of MS symptoms is a critical and active area of research. This thesis aimed to bridge and aggregate data from Finnish multiple sclerosis register and Helsinki University hospital (HUS) patient data system to create an aggregate dataset which can then be used as a basis for further analysis. Furthermore, a machine learning model was developed to enrich the aggregate dataset with information about the lesions, based on classification from radiologists' statements. The model aims to classify, whether a statement contains remarks of new lesions compared to previous statements and thus acts as tool for automatic lesion detection in order to alleviate manual labour tasks. The machine learning model was trained on 1000 manually labeled radiologists' statements and it achieved a total classification accuracy of 89\%, indicating that machine learning models have significant potential as assistive tools for reducing manual labour. However, more work is required for developing more accurate and comprehensive models and for integrating them to existing data processing systems. From the enriched aggregate dataset, generalized linear models (GLM) were built to predict how disease relapses before diagnosis, lesions before diagnosis, expanded disability status scale (EDSS) progression, and demographic factors such as sex and age at the time of diagnosis predicted accrued relapses after two, five, and ten years. The colinearity of the independent variables were validated using variance inflation factor (VIF). Significant predictive power was observed for male sex and age at the time of diagnosis, showing a negative correlation with the number of relapses accrued during the first two years of the follow-up time. Furthermore, a higher number of relapses before diagnosis, as well as a higher number of predicted lesions indicated a lower number of relapses after diagnosis. The aggregation of the data limits the granularity of the data, and thus hides any temporal factors the data would contain. Furthermore, while the usage of GLM can predict which variablesinfluence or predict the relapses after diagnosis, the current models cannot give rise to causal mechanisms on how the variables affect the disability progression.
  • Evolution of public transport networks of 14 Finnish cities: efficiency, navigability, and regional differences
    (2024-09-26) Niskanen, Anni-Mari
     School of Science | Master's thesis
    A functional and efficient public transport system is essential for any modern city. This Master's thesis studies the public transport systems of 14 Finnish cities based on public transport timetable data. The systems and their changes were studied by inspecting and comparing two instances of each city's system, one from 2018 and another from 2024. Both the efficiency and navigability of the systems were examined. Furthermore, travelling to the public hospitals and health stations of the cities, critical locations that a public transport user might have to reach quickly and navigate easily to in an emergency, was inspected separately. It was also of interest to find groups of cities with similar public transport systems. The public transport systems of the cities were studied by presenting them as public transport networks (PTNs) and calculating a multitude of measures that describe the systems from the PTNs. The PTNs were modelled both as static and temporal PTNs, which omitted and included the PTNs' evolution in time, respectively. The PTNs were also clustered with hierarchical clustering by utilising some of the calculated measures. The results of this thesis show that most of the inspected PTNs rely mainly on routes that pass through the city centre. Moreover, almost all of the cities have invested in their PTNs to different extents between 2018 and 2024. Two PTNs have undergone overhauls whose effects are twofold, as some areas of the cities have benefitted from them while others have suffered. Conversely, the overhaul of one of the PTNs that has made it reliant on routes that pass through the city centre has mainly improved the PTN. Furthermore, some of the cities have heavily invested in their PTNs and made them more efficient overall, but at the cost of deteriorated navigability. Conversely, some of the other cities have made cuts to their PTNs, which has made the PTNs less efficient but more easily navigable in some areas of the cities. Additionally, it was discovered that most of the cities have neither particularly improved nor neglected the travelling to their public hospitals and health stations. Finally, two groupings of the PTNs were discovered, the first grouping separating the PTNs of large and small cities of Finland and the second separating the PTNs of cities in western and eastern Finland.