Machine learning-based statistical analysis for early stage detection of cervical cancer

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Cervical cancer (CC) is a leading cause of mortality in women globally, especially in developing nations.
• Early detection of CC significantly improves treatment outcomes.
• Existing diagnostic methods can be enhanced through advanced computational approaches.

Purpose of the Study:

• To identify efficient machine learning (ML) models for early-stage cervical cancer detection.
• To evaluate the performance of various ML algorithms using clinical data.
• To explore the impact of feature transformation techniques on classification accuracy.

Main Methods:

• Utilized a Kaggle cervical cancer dataset with four attribute classes: biopsy, cytology, Hinselmann, and Schiller.
• Applied three feature transformation methods: logarithmic, sine function, and Z-score.
• Assessed supervised ML algorithms including Random Tree (RT), Random Forest (RF), and Instance-Based K-nearest neighbor (IBk).
• Employed Feature Selection Techniques (FST) to identify key risk factors.

Main Results:

• Random Tree (RT) achieved 98.33% accuracy for biopsy and 98.65% for cytology data.
• Random Forest (RF) and Instance-Based K-nearest neighbor (IBk) showed high performance for Hinselmann (99.16%) and Schiller (98.58%) datasets, respectively.
• Logarithmic transformation was optimal for biopsy, sine function for cytology, and Z-score for Schiller data. Both log and sine functions performed best for Hinselmann data.

Conclusions:

• Machine learning models can accurately and efficiently detect early-stage cervical cancer from clinical data.
• System design, algorithm tuning, and appropriate feature transformation are crucial for optimal performance.
• This approach offers a promising tool for improving CC diagnosis and patient outcomes.