Improving prediction of cervical cancer using KNN imputer and multi-model ensemble learning

Area of Science:

• Computational biology and machine learning applications in oncology.
• Medical informatics and data science for disease detection.

Background:

• Cervical cancer remains a significant cause of mortality in women, necessitating advancements in early diagnosis and treatment.
• Automated cervical cancer identification using machine learning shows promise for improving diagnostic speed and accuracy.
• Missing data in medical datasets poses a substantial challenge, potentially compromising the performance of machine learning models.

Purpose of the Study:

• To address the critical issue of missing data in automated cervical cancer detection systems.
• To develop and evaluate a novel machine learning approach for robust cervical cancer identification.
• To compare the effectiveness of different imputation strategies (KNN Imputer, PCA) against data deletion.

Main Methods:

• A stacked ensemble voting classifier combining three machine learning models was developed.
• K-Nearest Neighbors (KNN) Imputer was employed to handle missing values within the datasets.
• Performance was evaluated across three scenarios: data deletion, KNN imputation, and PCA imputation.

Main Results:

• The proposed stacked ensemble model with KNN imputation achieved exceptional performance metrics.
• Achieved accuracy of 0.9941, precision of 0.98, recall of 0.96, and an F1 score of 0.97.
• KNN imputation demonstrated superior performance compared to data deletion and PCA imputation in this context.

Conclusions:

• The developed stacked ensemble model effectively handles missing data, significantly enhancing automated cervical cancer detection.
• This approach offers a powerful tool for medical experts, leading to more accurate cervical cancer therapy and improved testing.
• The research contributes to reducing the impact of cervical cancer on women's health and healthcare systems through improved diagnostics.