Parkinson’s Data Analysis

---

library(readr)
library(ggplot2)
library(tidyverse)

Importing the dataset

data <- read_csv("C:/Users/Admins/Desktop/parkinsons.csv")
data$name<- NULL

Boxplot of the ‘MDVP:Shimmer’ feature

ggplot(data, aes(x = factor(status), y = `MDVP:Shimmer`)) +
  geom_boxplot() +
  ggtitle("Boxplot of MDVP:Shimmer") +
  xlab("Status (0 = Healthy, 1 = Parkinson's)") +
  ylab("MDVP:Shimmer")+
  theme_bw()

# scatterplot of the ‘NHR’ vs. ‘HNR’ features

ggplot(data, aes(x = NHR, y = HNR, color = factor(status))) +
  geom_point() +
  ggtitle("Scatterplot of NHR vs. HNR") +
  xlab("NHR") +
  ylab("HNR") +
  scale_color_discrete(name = "Status", labels = c("Healthy", "Parkinson's"))+theme_bw()

# Barplot of the distribution of the ‘status’ feature

ggplot(data, aes(x = factor(status))) +
  geom_bar(fill= (col="darkgrey")) +
  ggtitle("Distribution of Status") +
  xlab("Status (0 = Healthy, 1 = Parkinson's)") +
  ylab("Count")+
  theme_bw()

# Density plot of the ‘D2’ feature

ggplot(data, aes(x = D2, color = factor(status))) +
  geom_density() +
  ggtitle("Density Plot of D2") +
  xlab("D2") +
  ylab("Density") +
  scale_color_discrete(name = "Status", labels = c("Healthy", "Parkinson's"))+theme_bw()

# histogram of “MDVP:Fhi(Hz)” column

ggplot(data, aes(x = `MDVP:Fhi(Hz)`)) +
  geom_histogram(fill = "blue", alpha = 0.5, binwidth = 30) +
  ggtitle("Histogram of MDVP:Fhi(Hz)") +
  xlab("MDVP:Fhi(Hz)") +
  ylab("Frequency")+theme_bw()

linear logistic model

# Load the caret library
library(caret)

# Load the necessary libraries
library(caret)

Split the data into training and testing sets

set.seed(123)
splitIndex <- createDataPartition(data$status, p = 0.7, list = FALSE)
train_data <- data[splitIndex, ]
test_data <- data[-splitIndex, ]

# Exclude the "name" column from the data
train_data_no_name <- train_data[,-1]
test_data_no_name <- test_data[,-1]

Fit a logistic regression model

logistic_model <- train(status ~ ., data = train_data_no_name, method = "glm", family = "binomial")

Predict the status on the test data

predictions <- predict(logistic_model, newdata = test_data_no_name)
# Evaluate the model performance using Mean Absolute Error (MAE)
mae <- mean(abs(predictions - test_data$status))
mae

## [1] 0.2018143

Linear regression model

# Load the necessary libraries
library(caret)

# Split the data into training and testing sets
set.seed(123)
splitIndex <- createDataPartition(data$status, p = 0.7, list = FALSE)
train_data <- data[splitIndex, ]
test_data <- data[-splitIndex, ]

# Exclude the "name" column from the data
train_data_no_name <- train_data[,-1]
test_data_no_name <- test_data[,-1]

simple linear regression model

model <- lm(status ~ ., data = train_data)

simple linear regression Model predictions

predictions <- predict(model, newdata = test_data)

# Evaluate the model performance using Mean Absolute Error (MAE)
mae <- mean(abs(predictions - test_data$status))
mae

## [1] 0.2984471

Knn linear model

library(class)

# Split the data into training and test sets
set.seed(123)
train_index <- sample(1:nrow(data), 0.7 * nrow(data))
train_data <- data[train_index, ]
test_data <- data[-train_index, ]

model_knn <- train(status ~ ., data = train_data, method = "knn")
summary(model_knn)

##             Length Class      Mode     
## learn        2     -none-     list     
## k            1     -none-     numeric  
## theDots      0     -none-     list     
## xNames      22     -none-     character
## problemType  1     -none-     character
## tuneValue    1     data.frame list     
## obsLevels    1     -none-     logical  
## param        0     -none-     list

# Predict on the test data
predictions_knn <- predict(model_knn, newdata = test_data)

# Calculate the MSE
mse <- mean((predictions - test_data$status)^2)
mse

## [1] 0.1618094

Report

The mean of the “MDVP:Fo(Hz)” variable, which represents the mean fundamental frequency, is significantly higher in individuals with Parkinson’s disease (127.7 Hz) compared to healthy individuals (115.2 Hz). This suggests that a higher mean fundamental frequency may be associated with Parkinson’s disease. Similarly, the mean of the “Jitter:DDP” variable, which is significantly higher in individuals with Parkinson’s disease (0.014) compared to healthy individuals (0.012). The mean of the “MDVP:Shimmer” variable, which represents the local variations in amplitude, is also significantly higher in individuals with Parkinson’s disease (0.066) compared to healthy individuals (0.055). This suggests that higher values of jitter and shimmer may be indicative of Parkinson’s disease.

To identify the best model between the Knn and the linear regression model, we used the The Mean Absolute Error (MAE). THe model with the lowest model is the best model for the data, therefore the KNN model is the best model with MAE of 0.1618094, followed by the linear logistic model with 0.2018143, and the linear regression model with a MAE of 0.2984471.

Conclusion : the Parkinson’s disease data set provides several insights into the characteristics of individuals with and without Parkinson’s disease. A higher mean fundamental frequency, higher values of jitter and shimmer, and positive associations between these variables and the presence of Parkinson’s disease suggest that they may be useful indicators of Parkinson’s disease. Further analysis and modeling is necessary to determine the strength of these associations and the utility of these variables in predicting Parkinson’s disease.