AI COMMONS HEALTH & WELLBEING HACKATHON SOLUTIONS
OVERVIEW
The AI Commons Project is a proof of concept of a new methodology of developing Artificial Intelligence solutions that allows anyone, anywhere to benefit from the possibilities that AI can provide. The project aims to increase/improve the accessibility, reproducibility, contextualization and enhancement of Artificial Intelligence solutions globally and especially in emerging markets.
The project aims to demonstrate how a global community of AI experts can learn and co-create mutually beneficial solutions with the opportunity for cross-county incremental enhancement.
HEART DISEASE PREDICTION
STATEMENT OF PURPOSE
INTRODUCTION
Data Science Nigeria
Ezekiel Adebayo Ogundepo (Data Science Nigeria)
Prof. W.B. Yahya (University of Ilorin)
Dr Olubayo Adekanmbi (Data Science Nigeria)
Rising Odegua (Data Science Nigeria)
David Olubukola Akanji (Data Science Nigeria)
PROBLEM DEFINITION
Heart disease refers to several types of heart conditions. According to research, it is one of the dominant diseases that affects people worldwide and often results to death. Medical research has identified that lifestyles, obesity, eating habits and physical inactivity are major factors that can lead to heart related diseases and other factors like smoking, high blood pressure, gender, ECG rate, cholesterol level, age, weight, height, hypertension can also increase the chances of heart diseases.
The World Health Organization estimates that about 17 million lives are lost yearly due to heart related diseases and that heart disease is prevalent in Asia, India and the United States. Also, research has shown that heart diseases are usually dominant in male than female and mostly affect middle aged and old people.
Various tests are carried out to diagnose heart disease. Tests results, medical history and past symptoms are examined by a doctor to determine whether a patient has a heart disease
Yes
Malaria detection using machine learning
Cancer detection using machine learning
SOLUTION
The solution used various patients’ health information and cardiovascular statistics to predict the likelihood of heart disease in patients. Ten (10) different classifiers and seven (7) evaluation metrics were used in the development of the solution and the first version was released in 2018
Article: Here
The output of the solution is a prediction of whether or not a patient has heart disease using the best performing model (SVM) out of the 10 trained models.
Health practitioners
Patients
Medical health personal: To provide historical data and context useful in training the models
Data scientists: Clean the data and perform various exploratory data analysis (EDA) before training the models and bring out analytical story telling to communicate the results to those that need it
Small amount of training data
More data should be used for training and evaluation
USAGE
The intended use of the solution is to predict if a patient has a heart disease. A sample use case would be- a patient goes to a hospital with symptoms of a heart disease, necessary patient’s information is taken and fed into the system, in seconds the system should accurately predict whether the patient has a heart disease.
Medical personnel
The input involves measurements on patients’ health and cardiovascular statistics as shown below:
Input Description
Age – age in years
Sex – sex of the patients
Cp – chest pain type
Trestbps – resting blood pressure (in mm Hg on admission to the hospital)
Chol – serum cholestoral in mg/dl
Fbs – fasting blood sugar > 120 mg/dl
Restecg – resting electrocardiographic results
Thalach – maximum heart rate achieved (beats per minute)
Exang – exercise induced angina
Oldpeak – ST depression induced by exercise relative to rest, a measure of abnormality in electrocardiograms
Slope – the slope of the peak exercise ST segment, an electrocardiography read out indicating quality of blood flow to the heart
ca – number of major vessels colored by flourosopy
Thal – results of thallium stress test measuring blood flow to the heart
The input is supplied to the system and the output of whether the patient whose information is supplied has a heart disease or not is displayed on the screen.
DOMAIN AND APPLICATIONS
DATASET
This study used two independent sources of dataset. The first one was obtained from a study of heart disease from Cleveland database that has been open to the public at the UCI (University of California, Irvine C.A) Machine Learning Repository and the second one was obtained from DRIVENDATA. Both datasets have various measurements on patients’ health and cardiovascular statistics to predict whether or not a patient has heart disease.
The UCI dataset was created by V.A. Medical Center, Long Beach and Cleveland Clinic Foundation: Robert Detrano, M.D., Ph.D.
Donor:
David W. Aha (aha ‘@’ ics.uci.edu) (714) 856-8779
COMPOSITION
The instances represent patients’ health and cardiovascular statistics for patient with or without heart disease.
The UCI heart disease dataset contains 13 features and 303 records
The DATADRIVEN heart disease dataset contains 13 features and 180 records
Each instance of both dataset contain medical information for patient with or without heart disease
Yes. The class of label in each dataset indicate present or absent of heart disease.
No.
COLLECTION PROCESS
PREPROCESSING/CLEANING/
LABELLING
USES
Yes
Relevant Papers that cited the UCI data:
Detrano, R., Janosi, A., Steinbrunn, W., Pfisterer, M., Schmid, J., Sandhu, S., Guppy, K., Lee, S., & Froelicher, V. (1989). International application of a new probability algorithm for the diagnosis of coronary artery disease. American Journal of Cardiology, 64,304–310.
David W. Aha & Dennis Kibler. “Instance-based prediction of heart-disease presence with the Cleveland database
Gennari, J.H., Langley, P, & Fisher, D. (1989). Models of incremental concept formation. Artificial Intelligence, 40, 11–61.
MAINTENANCE
UCI and Drivendata.
DATASET PUBLICLY AVAILABLE
Heart Disease Dataset from Cleveland database. Data description can be found in the link
Machine Learning with a Heart Dataset from DRIVENDATA. Data description can be found in the link
MODEL
MODEL DETAILS
Model date: 2019. Different classifiers, which included Logistic regression, random forest, Decision Tree (DT), Naive Bayes (NB), K-Nearest Neighbour (K-NN), Support Vector Machine (SVM), XGBTree, CForest, Linear Discriminant Analysis (LDA), and Artificial Neural Network (ANN) were implemented using caret package in RStudio.
We created a one-hot encoding of our categorical features and rescale numeric features in the dataset so they can have a similar range of values. We also apply a simple backwards feature selection also known as recursive feature elimination (RFE) using random forest on the dataset.
No.
Models such as KNN, SVM, and neural network required a transformation of features to be centered and/or scaled before being used in such models.
EVALUATION
TESTING THE SOLUTION
The data was split into train and test in the ratio 80:20 respectively and the performance was evaluated using seven (7) evaluation metrics which includes Accuracy, Sensitivity, Specificity, Precision, F1 score, Area under a ROC curve (AUC), and LogLoss.
TESTING BY THIRD PARTIES
RESULT
RESULT DETAILS
SVM outperforms all other classifiers used. It achieved an accuracy: 83.89%, Sensitivity: 82.5%, Specificity: 85%, Precision: 81.5%, AUC: 90.5%, LogLoss: 0.379, and F1 score: 82%. Several tuning of hyper parameters of models were done with caret package with 10- fold cross-validation to improve the performance of the models.
60 instances.
Evaluation metrics included Accuracy, Sensitivity, Specificity, Precision, F1 score, Area under a ROC curve (AUC), and LogLoss.
ENVIRONMENT
We used Windows 10 on 64 bit and statistical analysis was done in RStudio with R version 3.6.3. Packages used included rmarkdown for running the codes chunk by chunk, tidyverse for data analysis and visualization, and caret for classification training.
STEPS TO REPRODUCE THE SOLUTION
RESULT DETAILS
To reproduce the soluton:
Download and install R and Rstudio
Clone or download the project from https://github.com/gbganalyst/Heart-disease-paper
Run the file Heart_disease_journal.Rproj
SAFETY
GENERAL
Yes. There is possibility of misclassification by the solution but the error is minimized.
EXPLAINABILITY
FAIRNESS
CONCEPT DRIFT
SECURITY
The solution is not suitable for any project not related to heart disease classification.
User data is not collected.
This project is brought to you by
