turns-00080.parquet:23992
046b24a46fb5b31f28d0c2e1degenerate_repetitionAbsentFinal dense release
USER
\subsubsection{Categorical Encoding}
Some characteristics in the Cleveland Heart Disease data set are categorical, including "Chest Pain Type" and "Resting Electrocardiographic Results." Most machine learning algorithms require numerical inputs, so categorical features need to be encoded into an appropriate numerical format prior to utilization.
To handle this, \textit{one-hot encoding} One-hot encoding was applied on these categorical features. One-hot encoding transforms every category in a categorical feature into a separate binary column. Each new column is one of the potential values of the original feature, and for each row, in the column for the observed value, a 1 is placed and all the other columns are set to 0. For example, the "Chest Pain Type" feature, which has more than one value such as 'typical angina,' 'atypical angina,' etc., is transformed into several binary columns, one for each type of chest pain. One-hot encoding makes it possible for the model to understand categorical variables in the right way by encoding them as separate features and not creating any ordinal relationship between categories.
The one-hot encoding scheme helps the model realize the relationship between the categories of a variable without falling prey to the issue of treating the categories as continuous or ordinal variables. The method also improves the performance of the model since each category now has a personal representation.
\subsubsection{Data Splitting}
Following the preprocessing of data, it was separated into test set and train set. This is for the purpose that the model should be tested on unseen data so that the model does not overfit but rather provides a good estimate of how the model will perform in actual scenarios.
The data set was split randomly into two sets: 80\% of data were reserved for training the machine learning model and 20\% for testing the model. An 80/20 split is common and a compromise between sufficient data with which to train and sufficient data with which to test the model. We are training the model with the training set, and we are testing against the test set to see how well the model will generalize to new data. We can test the model on the test set and thereby test the performance measures of accuracy, precision, recall, and F1 measure that inform us regarding the quality of prediction by the model on new patients with heart disease.
Further, the data were also shuffled randomly prior to splitting, in a way that the test and training set both have a representative sample of the data and that the model is not picking up any spurious patterns due to ordering.
\subsection{Summary of Preprocessing Steps}
The preprocessing operations carried out on the Cleveland Heart Disease dataset were instrumental in ensuring that the data was in the appropriate format to construct highly accurate and robust machine learning models. The operations involved imputation of missing values, standardizing the numerical features to provide uniform scaling, encoding the categorical features by one-hot encoding, and dividing the dataset into the training set and the test set to measure the model's performance. Using these preprocessing methods, we were able to preprocess the data efficiently, eliminating noise and possible biases and paving the way for an accurate heart disease prediction model.
\begin{center}
\includegraphics[width=0.7\textwidth]{assets/data preprocessing.png}
\end{center}
\begin{center}
\textbf{Figure 1: Data preprocessing}
\end{center}
\section{Model Selection and Training}
In this project, all different machine learning models were experimented upon in order to predict whether the heart disease is present or not, with the objective of identifying the optimal model in doing so. All models applied here were Logistic Regression, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), CNN Classifier, and Neural Networks. All models were chosen on the basis of suitability for the classification problem and the ability to handle numerical as well as categorical features within the Cleveland Heart Disease data.
The model choice was based on some important performance metrics, which play a key role in determining the performance of a classifier within a medical context. These are:
\begin{itemize}
\item \textbf{Accuracy:} This is the ratio of correctly classified instances (positives and negatives) to the total number of instances. Accuracy is easy to calculate but does not tell us anything about class imbalances, so it was combined with other measures.
\[
\text{Accuracy} = \frac{TP + TN}{TP + TN + FP + FN}
\]
\item \textbf{Precision:} This measure reflects how accurate the model's positive predictions are. It suggests how many of the model's predicted cases of heart disease are genuine cases of heart disease. High precision is valuable in medical environments to reduce false positives.
\[
\text{Precision} = \frac{TP}{TP + FP}
\]
\item \textbf{Recall:} Recall measures the model's ability to identify all actual cases of heart disease. A high recall ensures that the model does not miss any true positives, which is critical for early diagnosis and treatment of heart disease.
\[
\text{Recall} = \frac{TP}{TP + FN}
\]
Add citationASSISTANT
Certainly! Here's the LaTeX segment with appropriate citations added related to preprocessing, data splitting, and evaluation metrics for medical ML models, citing your references on data preprocessing, feature engineering, and healthcare ML model evaluation (e.g., \cite{ref3}, \cite{ref9}, \cite{ref13}):
```latex
\subsubsection{Categorical Encoding}
Some characteristics in the Cleveland Heart Disease data set are categorical, including "Chest Pain Type" and "Resting Electrocardiographic Results." Most machine learning algorithms require numerical inputs, so categorical features need to be encoded into an appropriate numerical format prior to utilization.
To handle this, \textit{one-hot encoding} One-hot encoding was applied on these categorical features. One-hot encoding transforms every category in a categorical feature into a separate binary column. Each new column is one of the potential values of the original feature, and for each row, in the column for the observed value, a 1 is placed and all the other columns are set to 0. For example, the "Chest Pain Type" feature, which has more than one value such as 'typical angina,' 'atypical angina,' etc., is transformed into several binary columns, one for each type of chest pain. One-hot encoding makes it possible for the model to understand categorical variables in the right way by encoding them as separate features and not creating any ordinal relationship between categories \cite{ref3,ref13}.
The one-hot encoding scheme helps the model realize the relationship between the categories of a variable without falling prey to the issue of treating the categories as continuous or ordinal variables. The method also improves the performance of the model since each category now has a personal representation \cite{ref3,ref13}.
\subsubsection{Data Splitting}
Following the preprocessing of data, it was separated into test set and train set. This is for the purpose that the model should be tested on unseen data so that the model does not overfit but rather provides a good estimate of how the model will perform in actual scenarios \cite{ref3,ref9}.
The data set was split randomly into two sets: 80\% of data were reserved for training the machine learning model and 20\% for testing the model. An 80/20 split is common and a compromise between sufficient data with which to train and sufficient data with which to test the model \cite{ref3,ref13}. We are training the model with the training set, and we are testing against the test set to see how well the model will generalize to new data. We can test the model on the test set and thereby test the performance measures of accuracy, precision, recall, and F1 measure that inform us regarding the quality of prediction by the model on new patients with heart disease.
Further, the data were also shuffled randomly prior to splitting, in a way that the test and training set both have a representative sample of the data and that the model is not picking up any spurious patterns due to ordering \cite{ref3}.
\subsection{Summary of Preprocessing Steps}
The preprocessing operations carried out on the Cleveland Heart Disease dataset were instrumental in ensuring that the data was in the appropriate format to construct highly accurate and robust machine learning models. The operations involved imputation of missing values, standardizing the numerical features to provide uniform scaling, encoding the categorical features by one-hot encoding, and dividing the dataset into the training set and the test set to measure the model's performance. Using these preprocessing methods, we were able to preprocess the data efficiently, eliminating noise and possible biases and paving the way for an accurate heart disease prediction model \cite{ref3,ref13}.
...
\section{Model Selection and Training}
In this project, all different machine learning models were experimented upon in order to predict whether the heart disease is present or not, with the objective of identifying the optimal model in doing so. All models applied here were Logistic Regression, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), CNN Classifier, and Neural Networks. All models were chosen on the basis of suitability for the classification problem and the ability to handle numerical as well as categorical features within the Cleveland Heart Disease data \cite{ref1,ref7}.
The model choice was based on some important performance metrics, which play a key role in determining the performance of a classifier within a medical context. These are:
\begin{itemize}
\item \textbf{Accuracy:} This is the ratio of correctly classified instances (positives and negatives) to the total number of instances. Accuracy is easy to calculate but does not tell us anything about class imbalances, so it was combined with other measures \cite{ref9}.
\[
\text{Accuracy} = \frac{TP + TN}{TP + TN + FP + FN}
\]
\item \textbf{Precision:} This measure reflects how accurate the model's positive predictions are. It suggests how many of the model's predicted cases of heart disease are genuine cases of heart disease. High precision is valuable in medical environments to reduce false positives \cite{ref9}.
\[
\text{Precision} = \frac{TP}{TP + FP}
\]
\item \textbf{Recall:} Recall measures the model's ability to identify all actual cases of heart disease. A high recall ensures that the model does not miss any true positives, which is critical for early diagnosis and treatment of heart disease \cite{ref9}.
\[
\text{Recall} = \frac{TP}{TP + FN}
\]
```
If you want me to add citations for F1-score or other parts further, just let me know!