Advancements in the field of AI and ML have had a great impact on the field of healthcare and automating medical assistance. Yet the underlying skepticism prevails in the usage of such technology in such a field as it has been handled by professionals of the field. Hence the advent of Explainable Artificial Intelligence (XAI) assumes significance in this context. The proposed work aims at tackling such a problem in the context of skin lesion classification using XAI techniques such as Grad-CAM to provide an understanding of the prediction made by the model which has been trained on the HAM-10000 dataset. Using data augmentation techniques the Mobile Net model has been trained leading to better results than the original dataset in early detection of the seven conditions of skin lesions. Using Grad-CAM the image is visualized with an overlay of the heat map resulting in the prediction of the model hence aiding in understanding the result better and generating a factor of trust towards the machine’s predictions.

This paper presents two hyper-heuristic approaches for solving a facility location problem called reliable p-median problem with at facility service (RpMF). In RpMF, service is provided to customers at the facility locations and it is closely related to the p-median problem. p-median problem is concerned with locating p-facilities while minimizing the total distance traveled by the customers to the corresponding nearest facilities and it is an NP-hard problem. But according to the p-median problem, it doesn’t consider the possibility of facility failures. On the other hand, RpMF assumes that facilities can fail and the customers assigned to that facility do not know about the facility failure till they reach the facility for service. So, the customers have to travel from failed facilities to other functioning facilities to receive service. RpMF deals with locating p facilities to minimize the cost of serving the customers while considering facility failures. We have proposed two multi-start hyper-heuristic based approaches that are based on greedy and random selection mechanisms to solve the RpMF. The solutions obtained through hyper-heuristics are improved further via a local search. The two proposed hyper-heuristic approaches are evaluated on 405 RpMF benchmark instances from the literature. Experimental results prove the effectiveness of the proposed approaches in comparison to the state-of-the-art approaches available in literature for the RpMF.

Traffic rules violation is a serious crime that can also lead to accidents if not followed. There is a possible occurrence of human errors (corruption, unclear vision due to weather conditions) while monitoring the traffic, so Deep Learning models present a way to monitor traffic rules for two-wheelers if the driver doesn’t wear a helmet or triple rides. Within the context of this paper, we propose a model which can automatically detect and classify Indian vehicles from video recorded by a surveillance camera during frames extraction using the YOLOV3 (You Only Look Once) model which consists of Darknet53 as an architectural backbone that consists of 53 Convolutional Neural Network (CNN) layers. After performing object detection with vehicle classification, the model is supposed to segregate all the two-wheelers and perform helmet detection. For helmet detection, we proposed to build a custom-trained model from two variations of YOLO models—YOLOV3 proposed by AlexeyAB Darknet, and YOLOV8 proposed by Ultralytics using transfer learning. During the comparative analysis, we found that YOLOV3 provides a mean Average Precision (mAP) of 55.86% with an average precision of 82% for a confidence threshold of 0.25, and the latest updated version of the YOLOV8 model which uses CSPDarknet53 produced an mAP of 96.09% with an average precision of 96.90% for SGD optimizer and initial and final learning rate of 0.01 and batch size of 8. After detecting helmets from two-wheelers, automatic number plate recognition is performed using a similar YOLOV3 model followed by image super-resolution using ESRGAN (Enhanced Super-Resolution Generative Adversarial Networks) followed by Optical Character Recognition (OCR) using the tool Pytesseract. The deep learning models trained for performing object detection and segmentation are developed using transfer learning methodology to enhance the performance of pre-trained YOLO weights files to perform detections with lesser computational costs.

Given a set of potential sites for locating facilities, the disruptive anti-covering location problem (DACLP) seeks to find the minimum number of facilities that can be located on these sites in such a way that each pair of facilities are separated by a distance which is more than R from one another and no more facilities can be added. DACLP is closely related with anti-covering location problem (ACLP), which is concerned with finding the maximum number of facilities that can be located such that all the facilities are separated by a distance which is more than R from each other. The disruptive anti-covering location problem is so named because it prevents the “best or maximal” packing solution of the anti-covering location problem from occurring. DACLP is a-hard problem and plays an important role in solving many real world problems including but not limited to forest management, locating bank branches, nuclear power plants, franchise stores and military defence units. In contrast to ACLP, DACLP is introduced only recently and is a relatively under-studied problem. In this paper, two intelligent optimization approaches namely genetic algorithm (GA) and discrete differential evolution (DDE) are proposed to solve the DACLP. These approaches are the first heuristic approaches for this problem. We have tested the proposed approaches on a total of 80 DACLP instances containing a maximum of 1577 potential sites. The effectiveness of the proposed approaches can be observed from the results on these instances.

Given a set of potential facility location sites along with a positive weight associated with each site as per its importance, the anti-covering location problem (ACLP) is about locating a set of facilities at some of these potential locations such that no two facilities are closer than a given distance from each other and sum of weights of chosen locations is as large as possible. This NP-hard problem has several important real-world applications such as telecommunications equipment siting, locating military units, locating franchise outlets, locating obnoxious facilities, forest management and DNA sequence matching. There are weighted and unweighted versions of ACLP. The unweighted version of ACLP is widely studied in the literature. However, the weighted version did not receive much attention despite several real-world applications. In this paper, we have proposed two evolutionary approaches based on genetic algorithm (GA) and discrete differential evolution (DDE) to solve the weighted version of the ACLP. The proposed approaches are used to solve the 80 ACLP instances with upto 1577 potential sites. Computational results show the effectiveness of our approaches.

Facility location models involve identifying locations for facilities that provide services to the customers which are also called demand points. The p-median problem is a facility location problem which deals with locating p facilities in such a way that the sumtotal of demand-weighted distances between each demand point and its respective closest facility is minimized. The p-median problem does not take into account the possibility of failure of the facilities, but rather considers that the facilities once located will always be available to serve the customers or demand points. But, in reality, some of the facilities may face unpredicted disruptions, thereby forcing the customers to seek services from other functioning facilities. The reliability p-median problem (RpMP) concerns with locating facilities which minimize the cost while considering the cost of facility failures. In this paper, we have proposed a hyper-heuristic approach with naive Bayes classifier for the RpMP and compared its performance on standard benchmark instances with two best performing approaches for RpMP, viz. a genetic algorithm and a scatter search approach available in the literature. The results show the effectiveness of our approach in terms of the solution quality.

This paper proposes a steady state genetic algorithm based approach for solving the obnoxious cooperative maximum covering location problem (OCMCLP) on a network. In cooperative coverage models, it is assumed that each facility emits a signal that decays over distance. At each demand point the cumulative signal strength received from all the facilities is calculated. A demand point is deemed to be covered if the total signal strength received by it is not less than a given threshold. All facilities contribute to the coverage of each demand point. Such models are different from the individual coverage models where the coverage of a demand point is decided by the single facility closest to that demand point. Given a graph with the set of demand points, the set of edges between these demand points, and the non-negative real weights associated with each demand point indicating the total demand at each point, the OCMCLP is concerned with locating p obnoxious (undesirable) facilities either at the demand points or along the edges in such a manner that maximizes the uncovered demand. The proposed genetic algorithm based approach makes use of crossover and mutation operators designed as per the characteristics of the OCMCLP. Solutions obtained through these genetic operators are improved further by a local search strategy. The performance of the proposed approach has been evaluated on the standard benchmark instances available in the literature. Computational results clearly show our proposed approach to be better in comparison to the existing state-of-the-art approaches for the OCMCLP.