Indian Journal of Science and Technology
Year: 2018, Volume: 11, Issue: 22, Pages: 1-10
V. Thavavel1 * and M. Karthiyayini2
1 College of Computer Engineering and Sciences, Prince Sattam bin Abdulaziz University, Al Kharj, Saudi Arabia; [email protected]
2 Department of Computer Sciences and Engineering, Sethu Institute of Technology, Kariapatti, Virudhunagar – 626115, Tamil Nadu, India; [email protected]
*Author for correspondence
College of Computer Engineering and Sciences, Prince Sattam bin Abdulaziz University, Al Kharj, Saudi Arabia; [email protected]
Objectives: Alzheimer's Disease (AD) is a chronic disease that eventually leads to death. Early diagnosis is expected to improve the patient survival. Evidence from literature indicates that best combination of biomarkers can help in early diagnosis. Feature selection techniques are proven as a workhorse in biomarker selection. This work aims to contribute a new hybrid feature selection framework based on ensemble learning for biomarker identification with an objective to improve the robustness of the final set of selected biomarkers. Methods and Analysis: The proposed framework employs Significance Analysis of Microarray (SAM) filter to initially select most relevant biomarkers. Then, the selected biomarkers are reselected in wrapper phase using heterogeneous voting based ensemble of classifiers to enhance the robustness of the final set of selected biomarkers. An extensive experiment was conducted to demonstrate the effectiveness of the proposed framework against the individual base learner in term of sensitivity and specificity. Finally, the statistical significance of the selected biomarkers was also verified by ANOVA test. Findings: The reported experimental results demonstrated improvement in AD diagnosis accuracy and proved the potential of the proposed framework in selecting the stable set of biomarkers for early AD diagnosis alternative to its base learners. The diagnosis accuracy level obtained by the identified set of biomarkers is over 87%. The higher area under ROC curve revealed the advantage of the identified biomarkers in discriminating AD patients from healthy control diagnosis with sensitivity and specificity of 93%. The p-value of ANOVA test results confirmed the significance of the identified biomarkers in early AD diagnosis. Overall, the identified robust set of biomarkers is expected to open a new pathway for early AD diagnosis. Improvements: Although clinical studies are needed to conform the findings, in the light of the results reported in this paper, it is evident that the proposed framework stands out to find the significant biomarkers for early AD diagnosis.
Keywords: Alzheimer’s Disease Early Diagnosis, Biomarker Identification, Filter-Wrapper Model, Greedy Search and Ensemble Classifiers, Hybrid Feature Selection, SAM Filter
Subscribe now for latest articles and news.