A bowel obstruction is a serious ailment that develops when a causative agent obstructs either the small or the large intestines. If there is blockage of the bowel; there will not be gas passing and it slows down food movement. Intestinal obstruction is currently the most prevalent surgical emergency worldwide, which continues to be a challenge for surgeons despite advancements in medical and surgical technology. Examination for clinical signs of bowel obstruction involves careful inspection of the cut-off point and the location where it happens. However conventional x-ray views lack anatomical clarity, detecting intestinal obstruction on a conventional radiograph is difficult. To solve this problem the researchers proposed Deep-shallow hybrid feature extraction techniques. In this study, we used Gabor, Gabor with GLCM, CNN, and Gabor-GLCM-CNN (Deep-shallow) as a feature extraction technique and we have examined SoftMax, SVM, and Ensemble learning (Bagging) as a classifier to develop human bowel classification model. Image preprocessing techniques like image normalization, filtering is applied. Moreover, to enhance the visual quality and clarity of the image, we have used Contrast Limited Adaptive Histogram Equalization (CLAHE). We used SVM, ensemble learning and SoftMax for classification of bowel obstruction as Normal, LBO, and SBO. The proposed Deep-shallow (Gabor-GLCM-CNN) features trained with SVM classifier and ensemble learners are compared with an end-toend CNN and Ensemble learners with deep-shallow feature outperforms the other. The experimental result shows that ensemble learner (Bagging) with multiple learners has better generalization ability than single learners like SVM. We have developed a custom end-toend CNN model using SoftMax classier and achieves 91.28% accuracy, developed to test the generalization ability of end-to-end CNN compared to Deep-shallow features. We evaluated the performance of Gabor, Gabor-GLCM, CNN, and Deep-shallow (CNNGabor-GLCM) feature extraction techniques by using a multi-class SVM classifier with RBF kernel achieves, 81.21%, 84.12%, 85.23%, 90.16%, respectively. Deep-shallow with ensemble learning technique using Bagging (Decision tree and Random Forest) achieves a test accuracy of 96.20% and 98.88% respectively. Deep -shallow feature with Bagging_RF ensemble learner achieves 98.88% accuracy and registered as the highest accuracy in this study. Key words: Deep-shallow, Ensemble learning, Bagging, SoftMax, SVM, RF, DT