Performance of machine learning algorithms for lung cancer prediction: a comparative approach

Area of Science:

• Environmental Health
• Medical Informatics
• Oncology

Background:

• Rising PM 2.5 aerosol levels correlate with increased lung cancer incidence and mortality.
• Lung cancer is often diagnosed late due to subtle early symptoms, necessitating improved diagnostic tools.
• Digital clinical records and machine learning offer potential for earlier and more accurate lung cancer detection.

Purpose of the Study:

• To evaluate the efficacy of twelve machine learning algorithms for early lung cancer prediction.
• To identify the most reliable machine learning models for diagnosing lung cancer based on clinical data.
• To compare predictive performance using patient symptoms and lifestyle habits.

Main Methods:

• Utilized clinical data including eleven lung cancer symptoms and two patient habits.
• Applied and compared twelve distinct machine learning algorithms.
• Results analyzed using classification metrics and heat map correlation.

Main Results:

• K-Nearest Neighbor (KNN) model demonstrated significant predictive power.
• Bernoulli Naive Bayes (BNB) model also proved highly effective for early detection.
• Both KNN and BNB outperformed other tested algorithms in accuracy.

Conclusions:

• Machine learning, particularly KNN and BNB, offers a reliable approach for early lung cancer prediction.
• Leveraging clinical data with advanced algorithms can improve diagnostic accuracy and patient outcomes.
• Further research can refine these models for clinical implementation in lung cancer screening.