Developing new machine learning ensembles for quality spine diagnosis

• Computer aided diagnosis system integrated with MAP and SPC.
• Proposed hybrid machine learning models.
• New design for reliable inference system with higher software reliability and quality.
• KDE used for first time in distinguishing patients.
• DOE optimization in machine learning.

This research work adduces new hybrid machine learning ensembles for improving the performance of a computer aided diagnosis system integrated with multimethod assessment process and statistical process control, used for the spine diagnosis based on noninvasive panoramic radiographs. Novel methods are proposed for enhanced accurate classification. All the computations are performed considering steep error tolerance rate with statistical significance level of 5% as well as 1% and established the results with corrected t-tests. The kernel density estimator has been implemented to distinguish the affected patients against healthy ones. A new ensemble consisting of Bayesian network optimized by Tabu search algorithm as a classifier and Haar wavelets as the projection filter is used for relevant feature selection and attribute’s ranking. The performance analysis of each method along with major findings is discussed using various evaluation metrics and concludes with propitious results. The results are compared to the existing SINPATCO platform that uses MLP, GRNN, and SVM. The optimization of machine learning algorithms is obtained using Design of Experiments scheme to achieve superior prediction accuracy. The highest classification accuracy obtained is 96.55% with sensitivity, specificity of 0.966 and 0.987 respectively. The objective is to enhance the software reliability and quality of spine disorder diagnosis using medical diagnostic system and reinforce the viability of precise treatment.