ICT5356 Principles of Artificial Intelligence Report Sample

ICT5356 Principles of Artificial Intelligence

You will research a goal, problem or task that could be solved using:

● Supervised machine learning; with

● A dataset that can be captured within a .csv file or similar that you are able to source; and with

● A simple machine learning algorithm (one of the algorithms that you will explore during Week 2)

In your application to SAINT, you will describe:

● The goal, problem or task

● The kind of data that will be used to train the machine

● An example of the kind of dataset that could be used (by creating a small sample yourself, or sourcing an appropriate sample online)

● The features that will be explored in the training

● The learning algorithm(s) that could be used to learn from the training data

● One or more working models trained on the dataset

Specifically, the machine learning model will be developed within the Orange software package. You are required to submit three items:

1. A document (500 words) that includes:

● A description of the goal, problem or task that you want to solve

● An explanation of how this goal, problem or task could be framed as a classification or regression task

● A training dataset that you will use to address this goal/problem/task - either link to an actual dataset, or provide a sample of data that you create

● An explanation of this dataset - the features included, the target variable, and how the data relates to the classification or regression task

● Any modifications or pre-processing you did for this training data, and why

● The machine learning algorithms you used for the goal/problem/task, and why

● Any hyperparameters you specified for your machine learning models

● What you achieved with the model - describe the predictions the model made on a small sample of data

2. The Orange project.

3. The model(s) developed.

Solution

Goal and Problem Description

This project’s purpose is to predict the occurrence of a heart attack using a supervised machine learning approach. CVDs are the number one killer all across the globe, and amongst them, heart attacks and strokes are the most common occurrences. The goal of this project for MBA assignment expert is to build a model for predicting the likelihood of a heart attack based on the dataset containing heart disease indicators. Through the use of machine learning methodologies, the researcher is optimistic of coming up with early diagnosis so that the life of individuals affected could be saved apart from cutting down the impact of heart diseases.

The Dataset

The dataset used in this study is obtained from kaggle and contains the summary data of more than 1300 people.

The dataset can be accessed from: <https://www.kaggle.com/datasets/bharath011/heart-disease-classification-dataset>

Figure 1: The Data Table
(Source: Self-Created)

Framing as a Classification Task

The study can also present the problem of heart attacks prediction as a classification problem. In this context, the machine learning model is trained to classify individuals into two categories: positive and negative, where “positive” means that the patient is likely to have a heart attack and “negative” means that the patient is unlikely to have one. The use of the binary classification method is appropriate since the characteristic dataset record and future attributes can be learned by the model depending on the patterns shown in the provided sample of a dataset (Ratra et al. 2020).

Figure 2: Orange Workflow
(Source: Self-Created)

Dataset Description

The dataset used in this project is sourced from Kaggle and contains 1,319 samples with nine fields: This is comprised of eight fields which are the input ones and one for the output. The input fields are:

• Age

• Gender (0 for Female, 1 for Male)

• Heart rate (impulse)

• Systolic BP (pressurehight)

• Diastolic BP (pressurelow)

• Blood sugar (glucose)

• CK-MB (kcm)

• Test-Troponin (troponin)

The dependent variable or output field is “class” indicating the occurrence of heart attack. The class variable is binary where the class ‘negative’ means no experience of a heart attack ever and class ‘positive’ means experience of a heart attack.

Data Preprocessing

While preparing the data, observations with missing values were omitted to avoid dubious results. This step is very important because the missing values can greatly harm the performance of the model (El-Hasnony et al. 2022). The target column was set to “class,” and all others were set as feature columns. This preprocessing is very effective in cleaning up the dataset before feeding it to a model.

Machine Learning Model and Hyperparameters

The adopted machine learning model for this task is a decision tree classifier. The model parameters are:

• Pruning: At least five in leaves and at least 10 in internal nodes; the maximum depth is 50.

• Splitting: Halt splitting when reaching 90% majority in the classification only.

• Binary trees: Yes

The decision tree model was selected because of the interpretability of the model and because it can be used to model numerical as well as categorical variables (Gárate-Escamila et al. 2020). The specified hyperparameters avoid overcomplicated tree structures with a large depth while also avoiding overly simple trees.

Model Performance

The model was evaluated using the "Test and Score" widget in Orange, with the following results:

• AUC: 0.982

• CA: 0.986

• F1: 0.986

• Precision: 0.986

• Recall: 0.986

• MCC: 0.970

Figure 3: Evaluation Results
(Self-Created)

Achievements with the Model

From the "Prediction" widget, on a small sample data, the scores obtained:

• AUC: 0.989

• CA: 0.987

• F1: 0.987

• Precision: 0.987

• Recall: 0.987

• MCC: 0.973

Figure 4: Model Performance Score
(Source: Self-Created)

The confusion matrix for the model is as follows:

• rue Negative: 506

• False Negative: 3

• False Positive: 14

• True Positive: 796

Figure 5: Confusion Matrix
(Source: Self-Created)

The decision tree model in the present study performed well in predicting the cases of heart attacks with an AUC of 0.989. The model yielded high accuracy of 0.987 and specificity, which are evident marks of the potential of the model to accurately identify potential heart attack patients.

References

El-Hasnony, I.M., Elzeki, O.M., Alshehri, A. and Salem, H., 2022. Multi-label active learning-based machine learning model for heart disease prediction. Sensors, 22(3), p.1184.

Gárate-Escamila, A.K., El Hassani, A.H. and Andrès, E., 2020. Classification models for heart disease prediction using feature selection and PCA. Informatics in Medicine Unlocked, 19, p.100330.

Ratra, R. and Gulia, P., 2020. Experimental evaluation of open source data mining tools (WEKA and Orange). International Journal of Engineering Trends and Technology, 68(8), pp.30-35.