DR. Timothy Oladunni

Occam Razor suggests that the best model is the one that retains a parameter because it has a significant effect on the fitness of the model and the removal of which increases variance. In the words of Isaac Newton "We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances."

We demonstrated that this is the best approach to hedonic pricing. In the light of the present criticism of p value as a reliable statistical tool, then most work on implicit pricing using hedonic regression confronts a serious debacle. Literature is full of earlier works validated using only p value.

We argue that since proper interpretation of the set of implicit prices in a hedonic regression determines the utility of a commodity.

Therefore, using unreliable instrument is counterproductive and defeats the theory of hedonic pricing. Going forward, we do not discard the use of p value, however, this study focuses on an Occam’s Razor approach of parsimony and plurality in the analysis and investigation of the hedonic pricing econometric concept.

We designed, developed and demonstrated using machine learning approach the impact of foreclosure sales on changes in standard sales. Our experiment was conducted using time series regression, correlation and Granger causality. Stationarity test for the experiment was based on Augmented Dickey-Fuller, Durbin-Watson and Portmanteau tests. The experiment suggests that there is a strong evidence of correlation relationship between foreclosure sales and house values. Therefore, we do not have evidence to reject the null hypothesis. However, our experiment did not establish any evidence of causality relationship between the two series

Our new Project every processes had become fragmented; meaning quality and service were inconsistent. This lack of standardization was adversely impacting operating costs, productivity and customer satisfaction. For several years now Payfast has worked strategically with innovations as a means of developing new solutions, products and services. In line with this vision, Success was approached to find new payments solutions to offer Payfast customers on their website, including open invoice and partial payments options.

• Client: Techs Soft
• Date: April 2018
• Service: Web Development

We empower students to use pattern recognition and machine learning tools for

1. data visualization,
2. data analysis,
3. feature selection and reduction for high dimensional datasets,
4. ensemble learning,
5. active learning,
6. transfer learning
7. reinforcement learning, and
8. adversarial learning.

The velocity, veracity, volumes and varieties of data explosions in the recent time has opened a new challenge to the traditional analytical methodologies. We create an environment for students to use state of the art tools for knowledge discovery from these datasets. Experiments are based on theoretical and empirical methodologies for an efficient algorithmic implementation.

The goal of this project is to design, develop and evaluate a deep learning predictive model to suggest the reason as to why an infant cried. Infants express their needs through the physiological response to an unfavorable condition or environment. An infant may cry because of hunger, sleep, wet diaper, or pain. It may also be because of pathological conditions due to asphyxia, deafness or autism. Experienced nursing mothers or care givers have been trained over the years on how to interpret these reactions. However, we find it an interesting topic to explore the pattern of the cries and the facial expressions of an infant using learning algorithms. This will help inexperienced mothers and pediatricians to provide adequate and efficient care for the infant child.

Software developers and programmers contribute on a regular basis to common repositories like github. The goal of this work is using deep learning approach to explore these repositories and extract useful information and patterns. Unlike other natural languages, source code has some sematic and synthetic structures that make the ubiquitous NLP approach unrealistic. Since the syntax nature of source code are very important, obviously, the NLTK which is very useful in NLP has some limitations. Therefore, one of the pre-processing stages of our work is using abstract syntax tree to retain the unique structure of source codes. Practical application of this project includes:

1. Verification of the authorship of a source code
2. Software forensics
3. Plagiarism detection
4. Copyright investigation

The project explores the unique structure of the indigenous acoustic structure of our languages. Using a deep learning approach, we develop a learning algorithm to analyze the differences of the patterns of different people in various parts of the world. Our model will help profile individuals to their geographical area based on their acoustic analysis. Application of this work include forensic investigation, speaker recognition and linguistic training.

In the recent time, the use of cloud computing has been popular all over the world. There are lots of benefits associated with the issue of this modern technology, however, there is a continued concern from the public about the security of information. The goal of this project is to develop a model whereby if Alice requests for computation of encrypted messages E(M_1)….E(M_n), from Bob. He computes the ciphertext using the function F{E(M1),…E(M_n )}. After the completion of the task, Bob sends an un-decrypted- computed message back to Alice.

Homomorphism provides two ways to solve the same problem and arriving at the same answer. Given two ciphertext values x and y, homomorphic algorithms provide a means whereby arithmetic operation on the plaintext produces the same result as the arithmetic operation on the ciphertext. As part of this project, we will develop an algorithm with practical implementation to e-voting.

This work focuses on an algorithmic investigation of the housing market spanning 11 years using the hedonic pricing theory. Hedonic pricing theory is an econometric concept that explains the market value of a differentiated commodity using implicit pricing.

One of the major criticisms of hedonic pricing methodology is its lack of specificity on the number of characteristic features. This is a major concern mostly when dataset span several years.

Dimensions of dataset may be large because of too many geographical submarket variables in an area under investigation.  Grouping housing submarkets into smaller clusters has a considerable impact on the size of our dataset without a significant reduction in the performance of the model.

This study investigates the properties of the brain electrical activity from different recording regions and physiological states for seizure detection. Neurophysiologists will find the work useful in the timely and accurate detection of epileptic seizures of their patients. We explored the best way to detect meaningful patterns from an epileptic Electroencephalogram (EEG). Signals used in this work are 23.6 second segments of 100 single channel surface EEG recordings collected with the sampling rate of 173.61 Hz.

The recorded signals are from five healthy volunteers with eyes closed and eyes open, and intracranial EEG recordings from five epilepsy patients during the seizure-free interval as well as epileptic seizures. Feature engineering was done using; i) feature extraction of each EEG wave in time, frequency and time-frequency domains via Butterworth filter, Fourier Transform and Wavelet Transform respectively and, ii) feature selection with T-test, and Sequential Forward Floating Selection (SFFS). SVM and KNN learning algorithms were applied to classify preprocessed EEG signal. Performance comparison was based on Accuracy, Sensitivity and Specificity. Our experiments showed that SVM has a slight edge over KNN.

The goal of VAPOC project is to find out reasons as to why the black community is disproportionally impacted during the coronavirus pandemic. We believe that a combination of factors is responsible for the African Americans’ susceptibility to the COVID-19. This study is based on pattern recognition, knowledge discovery and scientific principles. We hypothesize that preexisting conditions, type of employment, and access to healthcare among other factors have significant influence in the higher death rate of African Americans during the COVID-19 pandemic. To accomplish the above goal, the following 3 objectives are proposed; 1) Design, develop and evaluate a COVID-19 model to determine vulnerability to coronavirus, 2) Development of a visualization and interaction tool to analyze COVID-19 patients’ dataset in an immersive and non-immersive environment, and evaluate how graphical objects (such as data-shapes) developed in accordance with the user’s requirements can enhance situational awareness, and 3) Design, develop and evaluate a deep learning model to predict extent of COVID-19 damage to discharged patients.

Early and accurate identification of the balance deficits could reduce falls, in particular for older adults, a prone population. Our work investigates deep neural networks’ capacity to identify human balance patterns towards predicting fall-risk. Human balance ability can be characterized based on commonly used balance metrics, such as those derived from the force-plate time series. We hypothesized that low, moderate, and high risk of falling can be characterized based on balance metrics, derived from the force-plate time series, in conjunction with deep learning algorithms. Further, we predicted that our proposed One-One-One Deep Neural Networks algorithm provides a considerable increase in performance compared to other algorithms. Here, an open-source force-plate dataset, which quantified human balance from a wide demographic of human participants (163 females and males aged 18-86) for varied standing conditions (eyes-open firm surface, eyes-closed firm surface, eyes-open foam surface, eyes-closed foam surface) was used. Classification was based on one of the several indicators of fall-risk tied to the fear of falling: the clinically used Falls Efficacy Scale (FES) assessment. For human fall-risk prediction, the deep learning architecture implemented comprised of: Recurrent Neural Network (RNN), Long-Short Time Memory (LSTM), One Dimensional Convolutional Neural Network (1D-CNN), and a proposed One-One-One Deep Neural Network. Results showed that our One-One-One Deep Neural Networks algorithm outperformed the other aforementioned algorithms and state-of-the-art models on the same dataset. With an accuracy, precision, and sensitivity of 99.9%, 100%, 100%, respectively at the 12th epoch, we found that our proposed One-One-One Deep Neural Network model is the most efficient neural network in predicting human’s fall-risk (based on the FES measure) using the force-plate time series signal. This is a novel methodology for an accurate prediction of human risk of fall.

Abstract — COVID-19 pandemic that broke out in the late 2019 has spread across the globe. The disease has infected millions of people. Thousands of lives have been lost. The momentum of the disease has been slowed by the introduction of vaccine. However, some countries are still recording high number of casualties. The focus of this work is to design, develop and evaluate a machine learning county level COVID-19 severity classifier. The proposed model will predict severity of the disease in a county into low, moderate, or high. Policy makers will find the work useful in the distribution of vaccines. Five learning algorithms were trained and evaluated. Class imbalance was addressed using under-sampling of the majority classes. The result of our experiment shows that the ensemble models outperformed the non-ensemble models by a considerable margin.

Abstract—SARS-Cov-2 is not to be introduced anymore. The global pandemic that originated more than a year ago in Wuhan, China has claimed thousands of lives. Since the arrival of this plague, face mask has become part of our dressing code. The focus of this study is to design, develop and evaluate a COVID-19 fatality rate classifier at the county level. The proposed model predicts fatality rate as low, moderate, or high. This will help government and decision makers to improve mitigation strategy and provide measures to reduce the spread of the disease. Tourists and travelers will also find the work useful in planning of trips. Dataset used in the experiment contained imbalanced fatality levels. Therefore, class imbalance was offset using SMOTE. Evaluation of the proposed model was based on precision, F1 score, accuracy, and ROC curve. Five learning algorithms were trained and evaluated. Experimental results showed the Bagging model has the best performance.

Abstract—In this work, an automatic detection algorithm for hypertrophic cardiomyopathy (HCM) is presented. Of particular interest is the algorithm’s ability to differentiate HCM subjects and healthy volunteers from a single lead ECG dataset. Suspected HCM subjects are identified by the primary clinical abnormality associated with HCM, left ventricular hypertrophy (LVH). In total, n=43 human subjects ECG datasets are investigated: n=21 healthy volunteers and n=22 left ventricular hypertrophy (LVH) patients. Significant differences of p-value 0.01 and 0.04 were found for the respective ECG parameters, S-wave amplitude and ST-segment, when differentiating between the LVH patients and healthy human volunteers.

In 2019 there was an outbreak of coronavirus pandemic also known as COVID-19. Many scientists believe that the pandemic originated from Wuhan, China, before spreading to other parts of the globe. To reduce the spread of the disease, decision makers encouraged measures such as hand washing, face masking, and social distancing. In early 2021, some countries including the United States began administering COVID-19 vaccines. Vaccination brought a relief to the public; it also generated a lot of debates from anti-vaccine and pro-vaccine groups. The controversy and debate surrounding COVID-19 vaccine influenced the decision of several people in either to accept or reject vaccination. Because of data limitations, social media data, collected through live streaming public tweets using an Application Programming Interface (API) search, is considered a viable and reliable resource to study the opinion of the public on Covid-19 vaccine hesitancy. Thus, this study examines 3 sentiment computation methods (Azure Machine Learning, VADER, and TextBlob) to analyze COVID-19 vaccine hesitancy. Using the three sentiment computation methods and five learning algorithms (Random Forest, Logistic Regression, Decision Tree, LinearSVC, and Naïve Bayes) with different combination of three vectorization methods (Doc2Vec, CountVectorizer, and TF-IDF), we built 42 different classifier models. In analyzing COVID-19 dataset, the result of our analysis suggests that compared with Azure Machine Learning and VADER, TextBlob is the most effective method. The study also shows that Covid-19 vaccine hesitancy slowly decreases over time; suggesting that the public gradually feels warm and optimistic about COVID-19 vaccination. Finally, the result of our experiment shows that TextBlob + TF-IDF + LinearSVC has the best performance in classifying public sentiment into positive, neutral, or negative with an accuracy, precision, recall and F1 score of 0.96752, 0.96921, 0.92807 and 0.94702 respectively. It means that the best performance was achieved when using TextBlob sentiment score, with TF-IDF vectorization and LinearSVC classification model. We also found out that combining two vectorizations (CountVectorizer and TF-IDF) decreases model accuracy.

This study investigates patients' electrocardiography (ECG) properties for an accurate prediction of COVID-19 infection. The findings will be helpful to medical practitioners in the prognosis of COVID -19. We analyzed datasets of patients who had tested positive to COVID-19 and Normal Persons who had tested negative. Using the analyzed dataset, we designed, developed and evaluated a machine learning model to predict COVID-19. The analyzed signals are from 276 Normal Persons and 139 COVID19 patients. Time, frequency, and time-frequency domain features were extracted using Time Series Feature Extraction Library (TSFEL). We deployed T-test to determine necessary and sufficient features for our predictive modeling. K-nearest neighbors (KNN), Random Forest, and Support Vector Machine (SVM) learning algorithms were used to classify the ECG signals for the prediction of COVID-19. Performance evaluation was based on accuracy, sensitivity, and specificity. Using the time-domain properties, our experiment showed 97.07% accuracy for Random Forest classification, 95.93% for SVM, and 94.19% for KNN classification. In the frequency domain, we achieved an accuracy rate of 97.33% for Random Forest, 97.1% for SVM, and 96.64% for KNN. In the Time-Frequency domain properties, the classification accuracy was 93.80% for Random Forest, 90.08% for SVM, and 71.49% for KNN. We also investigated a combination of domains' features comprising of Time, Frequency, and TimeFrequency attributes for prediction. The combined domains gave us an accuracy of 97.93% for Random Forest, 97.52% for SVM, and 98.34% for KNN. Our experimental results show that the classification encompassing all domain features (Temporal, Spectral, and wavelet) with KNN had the highest accuracy for predicting COVID-19 at 98.34%

The time series analysis shows the underlying patterns of the waves of COVID-19 in the states under investigation. The figure suggests that there is a disproportionality in the percentage of blacks who contracted COVID-19 to the percentage who died of it. For example, as of December 13, 2020, in FL, the total cases and black cases were 1,155,335 and 146,128 respectively- this is a 12.65% of black cases to the total cases. However, the total COVID-19 deaths and black deaths were 20,490 and 3,461 respectively- which is a 16.89% of black deaths to the total deaths. Also, in GA there were total cases and black cases of 488,338 and 132,709 respectively- this is a 27.18% black cases to the total cases. However, there were total deaths and black deaths of 10,228 and 3,567 respectively- this is 34.87% black deaths to the total deaths. In both FL and GA, the results suggest a disproportionality of 4.24% and 7.70% respectively.

As the daily number of COVID-19 pandemic cases changes in different parts of the world, there is a need to localize the elevated risk that accomplishes it. Therefore, this work focuses on a COVID-19 fatality prediction in each counties of the USA. Datasets from the coronavirus tracking of John Hopkins University was used for the experiment. Our study was based on COVID-19 deaths and cases from March to September 2020 in all the 3, 006 counties in the USA. Residents of different counties, business owners and travelers will find the work useful. The model also will help in the prioritization of vaccine distribution. Our machine learning model consists of ridge regression and neural networks. The ridge regression model is two folds; 1) a ridge regression model with t-test feature selection, and 2) a ridge regression model with M5 Prime feature selection. The neural network is also of three folds; 1) a shallow neural network, 2) a deep convolution neural network, and 3) a deep multiple layer perceptron. Performance evaluation of the models was based on the mean square error (MSE). The outcome of our experiment suggests that ridge regression with the t-statistics feature engineering has a slight edge over other models.

According to the Centers for Disease Control and Prevention (CDC), nursing homes are at high risk for infection and death from the novel coronavirus (COVID-19). In the United States, more than 143,000 elderly-care facility residents have been infected with the novel COVID-19. Within this population, it is estimated that 25,600 have died, accounting for a third of the country’s pandemic fatalities. Most COVID-19 patients were hypoxic or hypoxemic (SaO2 <90%) before intubation and during intubation. Additionally, many patients that go on to develop respiratory failure had hypoxemia but without any signs of respiratory distress, this is referred to as "silent hypoxemia". Also, patients have demonstrated silent hypoxia. Hypoxia and hypoxemia describe states of oxygen deficiency: hypoxia is deficiency in oxygenation at tissue or cellular level, while hypoxemia is a suboptimal normal partial pressure of oxygen. Hypoxia and hypoxemia are interrelated, as untreated hypoxemia will ultimately result in hypoxia. In this project, we propose to investigate the early detection of silent oxygen deficiency as an indicator of COVID-19 infection, in order to reduce the spread of COVID-19 in the elderly care living facilities.

The ultimate goal of this project is to determine the feasibility of detecting silent hypoxia/hypoxemia from the combined electrocardiogram (ECG) and respiratory rate (RR) parameters measured from a wearable sensor. The hypoxia detection algorithm utilize data obtain from a wearable device. The wearable device is equipped with a single-lead ECG sensor and tri-axial accelerometers. To obtain a respiratory signal from the tri-axial accelerometer measurements, a single, time varying and acceleration (linear or angular) vector is derived.

This work focuses on the spread and impact of COVID-19 in the black community. A detailed analysis will improve the universality of a comprehensive mitigation strategy in reducing and combatting the spread of coronavirus. Our analysis of COVID-19 spans March 2020 to November 2020. Forecasting computation was based on exponential smoothening. The quality of our models was evaluated using the Mean Absolute Percentage Error. Predominantly black states in the US are considered for the experiment. All things being equal, a forecast to February 2021 suggests a disturbing forecast for the African American communities in the states investigated.