Research studies indicate that skin temperature variations can be measured through the use of thermal images to detect diabetic foot ulcer infection. A number of authors have utilized thermal foot images to distinguish diabetic foot conditions characterized by sympathetic skin reaction from those without a sympathetic skin reaction [
14]. Furthermore, the findings of this study demonstrate thermal foot imaging’s efficiency in detecting the presence of diabetic foot diseases, monitoring their progression and evaluating the effectiveness of related treatments [
15]. A method of achieving this involves the division of images into six regions of interest (ROIs): arch, heel, forehead, lateral sole, hallux, and lesser toes, followed by the mean calculation of temperature differences for comparison with healthy participants’ temperatures. In a previous study, the authors proposed a system that utilized thermal images to diagnose diabetic foot conditions through the computation of corresponding point temperatures in both feet [
16]. The findings indicated that the average temperature among patients with neuropathy ranged between 32.8 and 27.9 °C. Additionally, in another study, the author presented a system for detecting diabetic foot ulcers, which entailed the application of image-processing techniques such as segmentation, registration, and abnormality detection for thermal images. The process of segmenting the foot image involved the utilization of active contours, while background removal was achieved through the application of B-spline non-rigid registration, resulting in alignment [
17]. The final step in the process involved the detection of abnormalities by subtracting the level of intensity in the corresponding position from the left foot to the right foot. Specifically, if the temperature difference exceeds a predefined threshold, it is deemed indicative of an abnormal region, as per the findings of this study.In another study, infrared images were employed for the discrimination of diabetic foot ulcer classes: those with no visible signs, those with local complications, and those with diffuse complications [
18]. The mean temperature difference between the ipsilateral and contralateral foot can be calculated to accomplish this. The researcher utilized thermal imaging to detect abnormalities in diabetic foot conditions by correlating area temperatures with color codes. The approach entailed employing a rainbow palette, which was segmented into 10 distinct colors based on differences between them [
19]. MATLAB mobile platform and Android smartphone were employed in the study to create an Android-based thermal system. Based on a comparison with a normal test image, the system predicted ulcers from four different ROI areas. Despite this, the study did not focus on cases that carried a high rate of complications, relying only on static thermal information in predicting ulcers in diabetics with grade-zero foot conditions [
20]. The use of infrared images for detecting diabetic foot conditions through the application of image processing techniques was explored in a previous study [
21], which reported the highest accuracy rate of 95.66%. However, the latitude of this research was limited to detect abnormalities. In the study conducted by [
22], they utilized an alternate convolutional neural network to classify diabetic foot ulcers. Meanwhile, in [
23], a thermal system was proposed to detect abnormalities in diabetic foot by extracting the textural and entropy features from the decomposed discrete wavelet transform (DWT) and higher-order spectra (HOS). However, the study was limited in that it could not distinguish between different types of diabetic foot conditions, achieving an identification rate of 89.39%. In [
24], the authors focused on developing an ensemble model that utilized a majority voting technique to combine unweighted predictions from various machine learning models. On the other hand, the authors of [
25] focused on discriminating techniques, where they combined the basic classifiers through a process that could adjust to the input observations and output requirements of each individual learning system.In the training set, the type of combination employed could be optimized by assigning weight to each classifier in order to improve the combined performance. In a study by the authors of [
26], ensemble-based methods were recommended as the most effective approach for data stream classification problems. Furthermore, the authors compared ensemble learning with sixty other algorithms. Similarly, the author of [
27] notes that in the realm of machine learning, multiple classifier systems have undergone considerable development in recent years, presenting promising solutions to a range of problems and enhancing precision.