Building trustworthy AI for healthcare through collaborative research and student mentoring
The mission of the Oladunni Lab is to create reliable and explainable AI systems that improve human health and decision making. We advance scientific understanding through innovative research, high-quality publications, and real-world impact. We also train students to become capable, ethical, and forward-thinking contributors to the computing field.
The Oladunni Lab conducts rigorous, theory-driven research in trustworthy AI for cardiovascular health. Our work is grounded in 9 complementary theoretical frameworks (CST, CFD, EAT, PIT, PECT, CPGT, ADT, AVCT, PTT) that span from foundational dynamics to clinical applications. We design methods that are interpretable, physiologically principled, and rigorously validated—combining deep learning, signal processing, and dynamical systems theory to solve problems in cardiac wellness monitoring, cuffless blood pressure estimation, and demographic equity in wearable AI.
The Oladunni Lab provides hands-on research experiences in machine learning, biomedical AI, and applied data science. Students work on real datasets, contribute to ongoing publications, and build practical skills that prepare them for graduate study and industry careers.
Genuine interest in AI, healthcare, or signal processing
Solid programming skills in Python, experience with ML libraries
Commitment to rigorous research and meeting deadlines
Ability to articulate ideas clearly in writing and presentations
Mentorship from experienced researchers
Hands-on experience with cutting-edge machine learning techniques
Opportunity to co-author peer-reviewed publications
Exposure to real medical datasets and research workflows
Professional development in academic writing and presentation
Interested? Contact Dr. Oladunni with your resume, CV, and a statement of research interests.
Implement the Cardiac Stability Index (CSI) from ECG and PPG signals using dynamical systems theory. Learn Lyapunov exponents, recurrence analysis, and entropy computation for physiologic complexity assessment.
Skills: Signal processing, nonlinear dynamics, Python/PyTorch
Design deep learning architectures that preserve complementarity across time, frequency, and time-frequency domains. Explore fusion strategies (early, late, hybrid) that enhance robustness without redundancy.
Skills: Signal representations, neural architecture design, TensorFlow/PyTorch
Link cardiac attractor geometry to hemodynamic outcomes. Build smartphone PPG-based cuffless BP estimators using single-calibration AVCT framework. Work with real wearable data and validate across diverse populations.
Skills: PPG processing, machine learning, mobile app integration, clinical validation
Develop Kalman state-space models for per-individual cardiac trajectory monitoring. Implement CUSUM drift detection to distinguish benign excursions from true physiologic changes in longitudinal data.
Skills: Kalman filtering, state-space modeling, time series analysis
Study interpretability methods for cardiac AI models, including saliency maps, attribution analysis, and uncertainty quantification. Validate that model explanations align with clinical reasoning and physiologic principles.
Skills: Model interpretability, visualization, clinical domain knowledge
Validate PPG-based algorithms across Fitzpatrick skin tones (1-6). Participate in IRB-approved longitudinal studies collecting diverse smartphone PPG data. Address demographic bias in wearable health monitoring.
Skills: Fairness metrics, bias detection, IRB protocols, diverse dataset collection
Student Performance Trends
Social Media Sentiment Analysis Using AI: Twitter Dataset
Accent Classification from Short English Speech
Traffic Sign Recognition for Self-Driving Car
Deep Learning in Vector Based Churn Prediction for Idle Games
Predicting Stock Prices Using Machine Learning
Breast Cancer Differentiation (Melaku, Kaleab, Surafel, Tena, Bruk)
Investigation of Balance Impairment Using Deep Learning
Banking Industry Stocks Analysis Using Econometric and Deep Learning
We maintain high standards for research quality, experimental design, and publication ethics. We aim to produce work that advances the field and withstands scrutiny.
We work together across disciplines, institutions, and perspectives. We value diverse viewpoints and collaborative problem-solving.
We stay current with the latest techniques, engage in professional development, and mentor the next generation of researchers.
We focus on problems that matter for human health and wellbeing. We design solutions with practical applicability and social responsibility in mind.
We believe AI systems should be understandable and trustworthy. We develop interpretable methods and communicate findings clearly to diverse audiences.
We conduct research ethically, respect data privacy, acknowledge limitations, and consider broader societal implications of our work.
Educating students is important for the advancement of society in creating new innovations and developing new technologies. Not only is student education important on a large, macroeconomic scale, it is also important in individual growth and advancement applied on a microeconomic scale for continuing economic stimulation and expansions in various industries. Developing a strong student performance, due to the large impact education can have on certain students' and opening up their opportunities in certain industries, is often important for a person's future career growth in addition to economic and social status, regardless of gender, age, income level, weight, or ethnicity.
My goal in this project is to use a twitter data set that contains a collection of tweets to detect the sentiment associated with a tweet and detect it as negative or positive accordingly using Machine Learning. Sentiment Analysis is the automated process of analyzing text data and sorting it into sentiments positive, negative, or neutral.
This project is designed to process continuous speech spoken by people from the various axis in the world and to differentiate among their specific languages that are included in the training dataset. The success of this algorithm could have a huge impact on the technology space by catering for different tonalities and accents in speech technology. It is important to note that while speech recognition technology has been used extensively to enhance the quality of living for individuals with disabilities, the motivation for the project highlighted in this paper is the former (controlling digital devices).
The purpose of this study is to accelerate the research to assist the vehicle in adopting AI and computer vision on the systems. The study is inspired by the concept of convolutional neural networks (CNN) using open source libraries like OpenCV and Keras. From the result, the model successfully identifies traffic sign, road lane, and adopt speed limits with very high accuracy. Applying AI on vehicles will play a more significant role in reducing road-related accidents and death.
Churn prediction estimates whether the user will continue to use a service or no longer use it. If a company can use this effectively, it can be used in various strategies for promotion. Particularly in the case of mobile games that use advertising-based revenue model as a business model. The number of users directly affects the profits, so it can be utilized more effectively. In this study, will examined the effectiveness of neural networks in churn prediction using real data from a mobile game that uses advertising-based revenue model. We experimented to see the effect of the neural networks on the churn prediction using mobile game's data. And although there were some differences in the experiment, it was confirmed that the overall neural network performed well. In this experiment, the algorithm with best performance was LSTM. Although only mobile game data is used here, the results of this study can be applied to other user-based businesses. In addition, the results obtained through such churn prediction can be used in various ways to improve the profits of the company.
Stock Market trading can be predicted through political events, news headlines, and general sentiments around companies in real time. We experimented prediction of stocks with machine learning algorithms. The Radial Basis Function being nonlinear had the best output that most resembled the shape of the data selected. The Linear graphs and Polynomial graph showed a linear curve and was mostly not able to pinpoint the prices. The LSTM function was able to grasp the shape of the graph but could not pinpoint any particular data points and was slightly off by some amount. The loss for this model was 0.0012 after 100 epochs. The MLP Classifier outputs a loss function which gives the amount of error handled in the running of the program. The minimum loss of the MLP was 0.00719235 after 51 epochs.
The goal of this project is to classify breast cancer using deep neural networks. We designed, developed and evaluated a convolutional neural network with hyperparameter tuning. Our model achieved a loss and accuracy of 0.07% loss and 97.3% respectively.
This work investigates the balance pattern of human subjects through the force-plate measurements. Four different 60s quiet standing tests of eyes-open/firm surface, eyes-closed/firm surface, eyes open/foam, eyes-closed foam performed by 163 participants in three trials. Totally 12 force-plate files were recorded for each subject with the sampling rate of 100 Hz. The dimension of data for each individual subject is thus 72000 rows by 9 columns. Different Deep Learning Algorithms were used in this work to discriminate between human subjects and detect balance impairment. MLP, CNN and RNN were used as learning algorithms with classification accuracy of 99.8%, 99.6% and 96.9% respectively.
Bank Industry stock prediction using LSTM
Analyzing stock prices is very difficult due to the uncertainty and their dynamic nature. Investors use many statistical techniques such as moving average, Relative Strength Index (RSI) and others to make buying or selling decision on stocks investment. Stocks price are times series data which is a model that deals with a sequence of measurement of a same variable based on series of time. In this paper, we evaluate the stocks using an econometric approach using granger causality which is a methodology used in econometric to analyse short- or long-term dependency in binary or multivariate data. Another technique of analysis that is also beneficial for investors is the prediction. The second of econometric used in the study is the Autoregressive Integrated Moving Average (ARIMA). Finally The study experiments a deep learning approach using the Long Short Term Memory(LSTM) which also cops with sequences of prediction using historical information and the Multilayer Perceptron (MLP) which is a class of feedforward artificial network. In this research, we conduct a predictive analysis of historical stock price data of three banking industry stocks of New York Exchange (NYSE) which are JP Morgan, Bank of America and Wells Fargo. We first evaluate their long-term relationship using their logarithm returns and follow by the prediction method. Based on the results obtained from the three methods meaning the LSTM, ARIMA and MLP, we conclude that LSTM outperforms ARIMA and MLP on the prediction.