2023 Proceedings

Permanent URI for this collection

https://ir.unilag.edu.ng/handle/123456789/13012

Browse

Browsing 2023 Proceedings by Title

Now showing 1 - 17 of 17
  • Item
    Open Access
    4G Base Station Placement using the Foraging Bees Optimization Algorithm
    (MIRG, 2023-10-28) Ebun F.; Idris D; Chukwuekwu O; Babatunde S; Chika Ojiako
    This comprehensive paper delves into the intricate challenge of enhancing wireless coverage in university campuses, with a particular focus on the dynamic environment of the University of Lagos in Nigeria. The primary aim is to provide pervasive wireless connectivity while ensuring energy-efficient operations. To accomplish this, we employ a Foraging Bee Optimization Algorithm (FBA) for strategically placing base stations. Using a fitness function that considers both wireless coverage and energy consumption, FBA identifies optimal base station locations. The results indicate substantial improvements in coverage, with 95% of the campus now within the network's reach compared to 87% using the Particle Swarm Optimizer (PSO). Nevertheless, the presence of uncovered areas and a power consumption penalty underscores the continuous need for further optimization and the integration of sustainable practices. Given the ever-evolving nature of wireless network optimization, this study underscores the significance of iterative approaches in maintaining optimal coverage and balancing energy efficiency.
  • Item
    Open Access
    A Comparative Study of LDA and PCA Feature Extraction Techniques for Content Based Movie Recommendation Systems
    (MIRG, 2024-10-28) Abdul, U.S.; Akingbehin, V.C; Oladejo, T.O; Fasina, E.P; Odumuyiwa, V.
    With the rapid growth of digital streaming platforms and the ever-expanding catalogue of movies, effective movie recommendation systems have become crucial to enhance user engagement and satisfaction. Content-based recommendation models offer a personalized approach by leveraging intrinsic movie features to suggest relevant films to users based on their preferences. This paper compares two content-based movie recommendation models that utilizes movie attributes to generate tailored movie suggestions. The proposed models based on Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) start by extracting and encoding various content-related features from movies, such as genre, director, actors, plot keywords, and textual summaries. These features form a high-dimensional representation of each movie, capturing its unique characteristics. A user's preferences and viewing history are then used to build a user profile, representing their movie tastes and preferences. By comparing the user profile to the encoded movie representations, the models can identify movies that match users’ preferences. This research contributes to the advancement of content-based movie recommendation models by showcasing the effectiveness of selected movie features in providing personalized and relevant movie suggestions to users. It also demonstrates that the LDA model outperforms the PCA model on the MovieLens dataset.
  • Item
    Open Access
    An Architecture for a Blockchain-Enhanced Federated Learning
    (MIRG, 2024-10-28) Iwegbulam, C.M.; Odumuyiwa, V.
    This research develops a decentralised machine learning technique for collaborative modelling that uses federated learning and blockchain technology to achieve model convergence. A blockchain-coordinated architecture is implemented enabling client nodes to train models on local private data and securely share model updates to build an aggregated global model. Extensive experiments validate the decentralized learning protocol on healthcare datasets. The federated models achieved convergence in root mean squared error metric at the sixth federated learning round by consolidating dispersed data. The results show that the secure multi-party computation method can be used to create accurate, combined models from data stored in different places and still achieve model convergence. However, there are overhead and scalability restrictions. This study lays the groundwork for decentralised machine learning with blockchain.
  • Item
    Open Access
    Borrowers' Loan Repayment Coefficient Prediction Using Machine Learning
    (MIRG, 2024-10-28) Mensorale, O.; Odumuyiwa, V.
    Nano loan and unsecured loans is a fast-growing trend in the Nigeria fintech space. User creditworthiness and loan default is a problem for the whole industry. Traditional credit scoring methods have limitations in assessing the creditworthiness of diverse applicants by Loan Credit officers. In most cases, businesses lose money due to high defaults in customer loans on their loan investment portfolios. The loan repayment coefficient is a measurement of how soon a user is likely to repay a loan based on the user's historical data or similar users in the same demographic data historical data. This paper presents an extensive analysis of data on 212,042 creditworthy loan applicants' repayment patterns. This pattern was explored and studied, finding key parameters like age, gender, loan amount etc. in building a machine learning model to predict a user loan repayment coefficient. Two machine learning algorithm was considered namely Linear Regression and Random Forest algorithm to run the prediction. The model was deployed and embedded into the application programming interface (API) which receives user data and computes the repayment coefficient. The computed repayment coefficient after using the machine learning models ranges from 0% to 100% where 100% is a perfect score implying the user will pay on time. The anticipated outcomes of this project are significant, promising a paradigm shift in the lending industry. It is expected to yield improved credit decisions and contribute valuable insights to the field of machine learning in credit assessment if adopted.
  • Item
    Open Access
    Categorizing Approaches to Justify Recommendations
    (MIRG, 2024-10-28) Yacouba Kyelem; T. Frédéric Ouedraogo; K. Kisito Kabore
    Recommendation justification enables users to understand the reasons and motivation behind the recommendation of an item in a recommender system. It makes the recommendation model much more transparent, and improves user satisfaction. It is because of the important role assigned to the justification of recommendations that the present work aims to identify the approaches and methods for justifying recommendations that exist in the literature. The state of the art has enabled us to categorize the different approaches to recommendation justification. There are two approaches to recommendation justification: the linked model and the post-hoc models. The data used for justification are external or internal to the items.
  • Item
    Open Access
    Design and Implementation of Amharic Text-to-Speech System for Visual-Impaired and Blind Students
    (MIRG, 2024-10-28) Walelign S.; Zewdie, E.; Yaregal, A; Shegaw, A.M.; Mastewal, M.
    This project focuses on the development of an advanced Amharic Text-to-Speech (TTS) system for visually impaired and blind students, with a primary emphasis on enhancing accessibility and usability. The comprehensive methodology encompasses Corpus Collection and Preprocessing, involving the assembly of a diverse Amharic language corpus and its meticulous preprocessing. Phonetic and Prosodic Modeling techniques are employed to capture the nuances of Amharic pronunciation. Additionally, the integration of Tacotron 2 and WaveGlow models, along with the training process, is detailed. The project extends its impact through the seamless integration of the TTS system into a mobile application, with a user-friendly interface designed specifically for visually impaired users. The anticipated outcome is a versatile and inclusive platform that empowers users to convert written text into spoken Amharic effortlessly. The success of the project is evaluated through extensive user testing, ensuring accessibility, usability, and naturalness in the synthesized speech for the targeted user group.
  • Item
    Open Access
    Development of an Obstacle-Avoidance Goal-Based Autonomous Vehicle using Arduino Uno and Infrared and Ultrasonic Sensors
    (MIRG, 2024-10-28) Joshua A. Gana; Ebun P. Fasina; Babatunde A. Sawyerr; Ramoni O. Yusuf; William J. Udousoro
    This project involves the advanced design and development of an intelligent goal-based obstacleavoiding robot. The autonomous robot in this project can locate a user command specifying a goal station on a track of stations while avoiding obstacles. The robot avoids collisions by altering its direction or course and then proceeding on a track or path. Based on the perception it has collected through the sensors, the robot decides which new path to take to avoid collisions and reach its goal. The robot is intelligent enough to follow tracks of black tape and stop when it reaches the goal. Its implementation employs an ultrasonic sensor to measure distances to obstacles, an IR (infrared) sensor to follow the laid-out tracks, and a Bluetooth module to allow it to receive commands from a user. It uses servo motors as actuators for the motion of its ultrasonic sensor and DC motor actuators for its wheels. The robot successfully followed and avoided obstacles on tracks in a layout that would allow it to reach shelves and pick and drop items in a warehouse.
  • Item
    Open Access
    Development of Models for Predicting California Bearing Ratio of Lateritic Soil Using Selected Soft Computing Techniques
    (MIRG, 2024-10-28) F.E Eze; T.E Adejumo; A. A Amadi; Yusuf A
    Soft computing approaches were used to create models that estimate the California bearing ratio values of lateritic soil. Soft computing techniques are algorithms that find provably correct and optimal solutions to problems. The Soaked CBR values used in pavement design take about 96 hours to complete the test process. This can be time-consuming and expensive, Hence the need for researchers to seek alternate means of obtaining it. Various studies have employed artificial intelligence techniques, including neural networks, genetic algorithms, and support vector machines, to estimate CBR values. While these approaches offer potential benefits, they also exhibit certain drawbacks, such as sensitivity to parameter settings, restricted adaptability, and difficulty in understanding the underlying relationships. This study proposes a new model to address this challenge, Artificial Neural Networks (ANN) and its hybrid (ANFIS) were considered. Soil samples were taken from a burrow pit, and the necessary testing was performed on the acquired soil samples. Index, compaction, and California bearing ratio tests were conducted. Two machine learning models, artificial neural networks (ANN) and adaptive neuro-fuzzy inference systems (ANFIS), were developed to predict CBR values of lateritic soil. The models were trained on 70% of the data and tested on the remaining 30%. Both models demonstrated satisfactory performance, but the ANFIS model exhibited superior accuracy, as evidenced by a higher R2 value (0.98), lower RMSE (0.11), and lower MSE (0.33). These results suggest that ANFIS is particularly effective in capturing complex relationships within the data and is a promising tool for predicting CBR values in lateritic soils.
  • Item
    Open Access
    Dynamic Terrain Generation With Deep GANs
    (MIRG, 2024-10-28) Kathan Gabani; Het Pathak; Isitbhai Thakkar; Nilesh Jain; Senthurapandi Rajendran; Dipyaman Mukherjee
    Exploring the realm of video game creation, procedural terrain generation has stood the test of time as a means to generate extensive new graphical content automatically. Conventional techniques often employ algorithms tailored for specific terrains, meticulously designed by human hands. This study delves into the innovative application of deep convolutional generative adversarial models (DC-GANs) for the dynamic fabrication of authentic terrain maps. Furthermore, we present an inventive methodology for feature extraction that facilitates the specification and manipulation of geographical attributes within the generated terrain.
  • Item
    Open Access
    Evaluating student’s performance using k- mean clustering
    (MIRG, 2024-10-28) Shahid Eqbal; Ayush Singh; Balkrishna Pandey; Deepa Singh
    The analysis and evaluation of a student's academic achievement are now difficult tasks for the academic community to do. Classifying student achievement is a difficult scientific task in the actual world. Thus, a system to evaluate students’ performance utilizing a deterministic model and the k-means clustering algorithm is described in this study. The analysis findings will help academic planners determine how well students performed over a particular semester and what initiatives they need to take to raise students’ performance.
  • Item
    Open Access
    Geo-parsing Locations during Natural Disasters for Emergency Intervention Management: A Case Study of Media Print Report of the Flood Incidence in Nigeria
    (MIRG, 2024-10-28) Sikiru, R.; Adekanmbi, O.; Soronnadi, A.; Akanji, D.
    In recent years, urban flooding has emerged as a severe and recurrent global natural disaster, leading to substantial human casualties and widespread infrastructure damage. Approximately, 23% of the world's population faces direct exposure to flood depths exceeding 0.15 meters during flood events, disproportionately affecting low- and middle-income countries. These floods result from a complex interplay of factors, including climate changes, urban development in flood-prone areas, sea-level rise, dam operations, and poor governance. The ability to quickly ascertain the specific locations affected by flooding is of paramount importance for alerting the public and enabling effective disaster response. Social media platforms and news outlets play a pivotal role in disseminating this critical information. This research introduces a comprehensive methodology tailored to identifying and visualizing flooded locations mentioned in online articles. By precisely identifying flooded places, emergency response teams will be able to allocate resources and aid those in need more efficiently during floods.
  • Item
    Open Access
    Geo-semantic profiling of brand-specific customer experience using citizen-generated social media comments
    (MIRG, 2024-10-28) Ngele, E.
    A good customer experience is likely to influence a customer’s decision to buy positively and equally a negative customer experience will most likely make a customer decide not to buy or go elsewhere. About 73% of customers said that one negative customer experience is enough for them to leave a brand, and go to a competitor. This research work will use text mining and sentiment analysis techniques to identify and categorize brand-specific social media comments into different groups, including positive, negative, and neutral comments. Geo-profiling techniques will be used to identify patterns and trends in customer feedback from different locations. The results will help businesses to understand how to use social media platforms effectively to collect and analyze customer feedback to improve their customer experience and also aid in providing tailored services and products to their customers based on their geographic location. This research will also contribute to the field of customer relationship management by providing new insights into how businesses can use geo-profiling techniques to enhance customer loyalty and satisfaction.
  • Item
    Open Access
    Geo-visualization of Hotspots of Citizens Dissatisfaction on Social Services Using Media Print: A Case Study of Fuel and Cash Scarcity in Nigeria
    (MIRG, 2024-10-28) Akojenu, S.; Adekanmbi,O.; Soronnadi, A.; Akanji, D.
    The recent public dissatisfaction in Nigeria due to cash and fuel scarcity underscores the critical role of these resources in the modern economy, impacting various aspects of society like transportation, commerce, and daily living expenses. Existing research on citizen dissatisfaction relied on surveys, but this study employs geosemantics techniques to extract locations of social service dissatisfaction from social media data for efficient resource allocation. The method involves crawling and classifying social media content into three categories (dissatisfied, satisfied, or neutral). Using deep learning and a rule-based geoparsing approach, the study identifies locations mentioned in dissatisfied text in real time. This real-time insight from unstructured text aids in comprehending the complex economic, social, and spatial effects of resource scarcity, facilitating the government in developing effective resource allocation strategies to improve citizens' quality of life.
  • Item
    Open Access
    Geo-Visualization of Road Accident and Traffic Congestion Hotspots for Real-Time Traffic Optimization, Mobility Planning, and Commuters’ Safety Using Traffic Reported Social Media Posts
    (MIRG, 2024-10-28) Lawal, O.; Adekanmbi,O; Soronnadi, A.; Akanji, D.
    Amidst the bustling economic hub of Nigeria, known as Lagos, traffic congestion remains a persistent challenge that plagues the city, impeding the economy, health, and safety of its residents. Professionals spend a staggering 90% of their work time navigating traffic, resulting in stress and anxiety. Some are robbed in the process, even in broad daylight. To address this problem, this work is done to help government agencies optimize traffic management and provide road users with the necessary information to navigate the city safely and efficiently and make informed decisions about which routes to take. In this study, we utilized LASTMA's social media tweets to analyse traffic patterns in Lagos, revealing insights on congestion and road safety. Rigorous data preprocessing and machine learning tools such as TfidfVectorizer aided sentiment analysis, achieving 88.9% accuracy with logistic regression outperforming other models. We also finetuned a name entity recognition (NER) model using Spacy to identify and capture location and road entities within the tweets. Findings highlighted breakdowns near sea ports and major routes (e.g., Orile, Bonny Camp, Apapa wharf) and accidentprone areas like the Third Mainland Bridge and Obalende Bridge. With this analysis, government agencies can develop traffic management strategies that prioritize safety and efficiency, such as rerouting traffic away from congested areas or implementing safety measures at accident-prone or traffic robbery-prone locations. Furthermore, traffic-free routes can be provided in real-time to commuters based on their location. This work has the potential to reduce traffic congestion and improve road safety, leading to economic and health benefits for Lagos residents. The results of this research not only have practical implications for the city but can also be extended to other urban areas that have traffic agencies and are also facing similar challenges.
  • Item
    Open Access
    Geosemantic Surveillance and Profiling of Abduction Locations and Risk Hotspots Using Print Media Reports
    (MIRG, 2024-10-28) Ogunremi, T.; Adekanmbi, O.; Soronnadi, A.; Akanji, D.
    Kidnapping poses a significant social risk in Nigeria, often exacerbated by the lack of local crime data, underreporting of cases, and potential involvement of security operatives. Our research aims to combat this menace by developing a data-driven solution that offers comprehensive insights into crime locations and entities. We have generated a reliable dataset by geoparsing newspaper-reported crime locations and entities using Natural Language Processing (NLP) techniques and Google geocoder. Additionally, we implemented clustering and geospatial analysis to identify social risk hotspots. Our method involves designing an algorithm that can geoparse locations in unstructured raw text. The results of our research provide crucial insights and solutions for addressing the threat of kidnapping in Nigeria. We recommend the implementation of our data-driven approach as an intervention strategy to aid law enforcement and policy makers. Our study contributes to the understanding of the spatiotemporal dynamics of kidnapping cases in Nigeria.
  • Item
    Open Access
    Kenyan Sign Language (KSL) Dataset: Using Artificial Intelligence (AI) in Bridging Communication Barrier among the Deaf Learners
    (MIRG, 2024-10-28) Wanzare, Z.; Okutoyi, J.; Kang'Ahi, M.; Ayere, M.
    Kenyan Sign Language (KSL) is the primary language used by the deaf community in Kenya. It is the medium of instruction from Pre-primary 1 to university among deaf learners, facilitating their education and academic achievement. Kenyan Sign Language is used for social interaction, expression of needs, making requests and general communication among persons who are deaf in Kenya. However, there exists a language barrier between the deaf and the hearing people in Kenya. Thus, the innovation on AI4KSL is key in eliminating the communication barrier. Artificial Intelligence for KSL is a two-year research project (2023-2024) that aims to create a digital openaccess AI of spontaneous and elicited data from a representative sample of the Kenyan deaf community. The purpose of this study is to develop AI assistive technology dataset that translates English to KSL as a way of fostering inclusion and bridging language barriers among deaf learners in Kenya. Specific objectives are: Build KSL dataset for spoken English and video recorded Kenyan Sign Language and to build transcriptions of the KSL signs to a phonetic-level interface of the sign language. In this paper, the methodology for building the dataset is described. Data was collected from 48 teachers and tutors of the deaf learners and 400 learners who are Deaf. Participants engaged mainly in sign language elicitation tasks through reading and singing. Findings of the dataset consisted of about 14,000 English sentences with corresponding KSL Gloss derived from a pool of about 4000 words and about 20,000 signed KSL videos that are either signed words or sentences. The second level of data outcomes consisted of 10,000 split and segmented KSL videos. The third outcome of the dataset consists of 4,000 transcribed words into five articulatory parameters according to HamNoSys system. The dataset is significant to members of the deaf community in breaking communication barriers among deaf and promoting wider inclusion of the deaf in education and other sectors. The scalability of the innovation is key in empowering the deaf worldwide.
  • Item
    Open Access
    Proposed Methodology on Enhancing Food Security in Nigeria Using Real Time Consumer Market Availability Application
    (MIRG, 2024-10-28) Akande O.; Adewuyi J.O; Adebanjo O.A; Adegbie F.M
    As part of United Nations Sustainable Development goals, food security is one of the five targets of goal number 2 which is zero hunger by the year 2030. To achieve this market needs to function properly and citizens need timely information about food availability in the food market. This will make people to visit the market at appropriate time, buy what they need in its most nutritious (quality) state, consume fresh foods and thus prevent spoiling and wastages of perishable foods in the market. This paper proposed an artificial intelligent based mobile app which bridged the gap between sellers of tomato and consumers such that tomato does not overstay and be spoilt in the market. The research methodology has three stages: image classification, the recommender system using model-based deep reinforcement learning, and deep learning production. Image acquisition for trained tomato is done using ordinary digital camera, noise and background were removed. Image segmentation was done using artificial neural network (ANN) to separate tomato images into segment. Next, feature extraction was also done using feature fusion-based extraction while image classification into good, manageable or bad will be done using object-based ANN. A buyer can give feedback, and the feedback will be trained to provide future recommendation to another buyer. This was done using deep reinforcement learning using the Markov property. Both the model and the recommendation system were integrated into mobile app to make it accessible to the public.