Dataset information

Row

Raw data’s feature

Feature	Description
sudden	Sudden Sensorineural Hearing Loss happen
ID	A unique identifier for each individual
sex	The biological sex
age	The age of the individual
hypertension	This individual has hypertension
dyslipidemia	This individual has dyslipidemia, which is an abnormal amount of lipids in the blood
stroke	This individual has stroke
MI/angina	This individual has Myocardial infarction
Diabetes	This individual has diabetes.
Thyroid	This individual has thyroid dysfunction
CRF	This individual has Chronic Renal Failure
Height	The height of the individual, in centimeters
Weight	The weight of the individual, in kilograms
BMI	Body Mass Index of the individual
HE_wc	Waist circumference, possibly measured at the health examination in unit of centimeters
TOTALchole	Total cholesterol levels in the blood
HDL	High-Density Lipoprotein cholesterol levels
TG	Triglycerides
LDLcal	Low-Density Lipoprotein cholesterol
Phospholipid	Phospholipid levels in the blood.
Free Fatty Acid	Free fatty acid levels in the blood.
Lipoprotein	Likely indicates the concentration of a certain lipoprotein subclass.
Lipid_total	Total lipid levels in the blood.
Followup_weeks	The number of weeks of follow-up after an initial event or diagnosis.
Siegel_criteria	Likely a scoring or classification system related to the study, perhaps to grade the severity of hearing loss.
Initial_audio(dB)	The level of hearing loss at first diagnosis.
Last_audio(dB)	The current level of hearing loss.
AUDIORt	An audio test result for the right ear
AUDIOLt	An audio test result for the left ear

Row

Cleaned feature used in this report

Feature	Description
sudden	Sudden Sensorineural Hearing Loss happen
sex	The biological sex
age	The age of the individual
hypertension	This individual have hypertension
dyslipidemia	This individual have dyslipidemia, which is an abnormal amount of lipids in the blood.
stroke	This individual have stroke.
MI_angina	This individual have Myocardial infarction
Diabetes	This individual have diabetes.
Thyroid	This individual have thyroid dysfunction
CRF	This individual have Chronic Renal Failure
Height	The height of the individual, in centimeters.
Weight	The weight of the individual, in kilograms.
BMI	Body Mass Index of the individual.
HDL	High-Density Lipoprotein cholesterol levels
TG	Triglycerides
LDLcal	Low-Density Lipoprotein cholesterol

Multiple Regression

TG Prediction Using Multiple Regression

Multiple regression allows us to use data from various columns to predict TG levels, which are typically measured through a blood test. If a reliable regression model is developed, it can save costs associated with additional testing. Our analysis using best subset regression suggests that a model with four variables (age, CRF, Weight, HDL) has the lowest BIC value and a relatively high adjusted \(R^2\). However, visual inspection of the model’s predictions compared to actual TG levels indicates the model may not be ideal for this dataset. Moreover, the regression line’s jagged appearance in the TG vs. HDL graph may suggest overfitting.

Row

Find Best Subset Regression Evaluation

Row

Multiple Regression Model for TG

Coefficients of the Multiple Regression Model for TG
	Estimate	Std. Error	t value	Pr(>\|t\|)
(Intercept)	121.622	17.627	6.900	0.000
age	0.499	0.139	3.588	0.000
CRF	93.973	31.132	3.019	0.003
Weight	0.890	0.180	4.945	0.000
HDL	-1.615	0.154	-10.498	0.000

Ridge Regression

Predict TG Using Ridge Regression & Log Transformation

In this algorithm, the magnitude of the coefficients derived from the Ridge Regression Model suggests a significant association between “TG” and the variables “CRF,” “dyslipidemia,” and “Thyroid,” as evidenced by their relatively large coefficients.

Upon examining the fitted regression line, evidence of heteroscedasticity was observed, characterized by non-constant variance in the residuals of the predicted values. To address this, a logarithmic transformation was applied, significantly improving the model’s fit. This improvement is quantitatively supported by the reduction in the Mean Squared Error (MSE), which decreased from 3248.609 to 72.087, underscoring the effectiveness of the logarithmic transformation.

Row

Ridge Regression: using all feature to predict TG

Coefficients of the Ridge Regression Model for TG
Term	Coefficient
(Intercept)	138.745
sudden	13.721
sex	-8.458
age	0.274
hypertension	5.976
dyslipidemia	22.123
stroke	8.175
MI_angina	5.920
Diabetes	8.316
Thyroid	-21.576
CRF	60.607
Height	-0.073
Weight	0.316
BMI	1.449
HDL	-1.259
LDLcal	-0.044

Row

Ridge Regression: using all feature to predict TG

Evaluation Value for Ridge Regression
Metric	Value
MSE	3248.609
R2	0.182
Adjusted_R2	0.149

Row

Ridge Regression: Log Tranformation on TG

Evaluation Value for Ridge Regression After Log Transformation
Metric	Value
MSE	72.087
R2	0.982
Adjusted_R2	0.981

natural cubic spline

Natural Cubic Spline to Predict TG Using HDL

For the natural cubic spline analysis, we initially evaluated the optimal degree of freedom by examining the “MSE for Different Degrees of Freedom” graph. This progress shown that a natural cubic spline with three degrees of freedom (df=3) is the optimal selection based on the lowest mean squared error.

After fitting the data with a natural cubic spline of df=3, the resulting fit line appeared to be well-balanced, demonstrating a smooth curve without signs of overfitting. However, the model exhibited a relatively high mean squared error (MSE) and a low coefficient of determination (\(R^2\)) for both in-sample and out-of-sample data. This outcome could be attributed to the dispersed nature of the data, suggesting that the relationship between the variables may not be readily observable.

Row

Scatter Plot for TG vs. HDL & Degree of Freedom Test.

Row

Scatter Plot With Natural Cubic Spline

Evaluation Value for Natural Cubic Spline
Metric	Value
in sample MSE	3892.0849
in sample R2	0.1586
out of sample MSE	3194.8315
out of sample R2	0.1956

kNN classifiication

Introduction

This study utilizes k-Nearest Neighbors (kNN) classification to investigate whether common health indicators—Body Mass Index (BMI), high-density lipoprotein (HDL), and triglycerides (TG)—can predict the risk of sudden sensorineural hearing loss (SSNHL). Through machine learning, we aim to determine the significance of these indicators in relation to SSNHL, optimizing our kNN model by testing various neighbor values. Our goal is to provide insights into potential early warning signs for SSNHL, which could be instrumental in guiding preventive measures.

Conclusion

The results from our kNN model highlight a clear link between lipid levels and the risk of SSNHL. The boxplot we’ve created shows small changes in SSNHL risk predictions when we adjust the number of neighbors (k) used in the model, with k=10 giving slightly higher risk estimates. On average, the risk predictions are 24.38% for k=3, 25.82% for k=5, and 25.07% for k=10. This suggests that higher levels of cholesterol and HDL are important to watch for as they could signal a higher chance of developing SSNHL. By incorporating kNN models into routine health evaluations, doctors could potentially improve how they predict and prevent SSNHL, aiming for better health outcomes.

Row

kNN classifiication

Row

kNN Table

KNN Classification Results for Different Values of k
k	True_Negatives	False_Positives	False_Negatives	True_Positives
3	158	57	35	8
5	163	57	30	8
10	180	60	13	5

Naive Bayes Classification

Introduction

Ｗe’re using Naive Bayes classification to sift through data on SSNHL, HDL, and BMI. This statistical approach will allow us to categorize each person into different risk levels for sudden hearing loss based on these health indicators. By analyzing the patterns in this data, Naive Bayes helps us identify who is at a higher risk. This insight is vital for early intervention, where we can offer specific guidance or treatment to those identified as high-risk, aiming to reduce their chances of experiencing sudden hearing loss.

Conclusion

The plot suggests a subtle trend where individuals aged 41-60 might face a higher risk of sudden hearing loss, as seen by a denser cluster of red dots, which represent SSNHL cases. This age group’s data points seem more tightly packed, indicating that factors contributing to SSNHL might be more prevalent or pronounced in this demographic. While the risk appears to span all ages, the middle-aged group stands out, hinting that factors like age-related health changes could play a role. Nonetheless, the blend of red and blue dots in every age category implies that SSNHL is likely a complex condition affected by multiple factors beyond just age and the two health measures of BMI and HDL displayed here.

Row

Naive Bayes Classification

Bayes Table

Performance Metrics
Performance Metrics
Metric	Value
Accuracy	0.7209302
Precision	0.7320000
Recall	0.9734043
F1 Score	0.8356164

Classification using Logistic Regression

Introduction

Utilizing logistic regression, this study targets the identification of individuals at heightened risk for sudden sensorineural hearing loss (SSNHL), with an emphasis on the roles of lipid profiles and obesity. This statistical approach allows for the differentiation between high-risk and low-risk groups based on key predictors such as total cholesterol, triglycerides, and BMI. The goal is to enable earlier interventions for those at risk and to contribute to the understanding of SSNHL’s risk factors, paving the way for targeted preventive strategies.

Conclusion

The ROC curve derived from logistic regression analysis, focusing on the correlation between lipid profiles, obesity, and the incidence of sudden sensorineural hearing loss (SSNHL), indicates that the model has good predictive power. The graph shows that as the body mass index (BMI) and lipid measures—specifically total cholesterol and triglycerides—escalate, so does the risk of SSNHL, corroborating the premise that vascular health is significantly tied to the onset of this condition. Consequently, these factors may serve as reliable indicators for identifying individuals at heightened risk for SSNHL, which can lead to earlier and potentially more effective interventions.

Row

Classification using Logistic Regression

Logistic Regression Table

Logistic Regression Classification Confusion Matrix
	Actual Class
	0	1
0	962	316
1	7	8

Logistic Regression Performance Metrics
Performance Metrics
Accuracy	Precision	Recall	F1.Score
0.75	0.53	0.02	0.05

Dataset information

Sidebar