Over the last several years, cloud computing has imposed as a major paradigm in software development by providing computer resources over the Internet. Among various cloud service models, Backend as a Service (BaaS) stands out as a model that targets the specific needs of web and mobile developers. By providing the backend for applications, it facilitates and expedites the software development process. In order to prevent major problems with the use of third-party BaaS providers, this paper advocates building your own BaaS platform, as well as several works ahead of it. However, the development of a BaaS platform carries various challenges regarding architecture and design. This paper strives to define the core service offerings of a BaaS platform and propose a method for providing an architectural design of a BaaS platform based on a microservice architecture. Microservice architecture is the preferred architectural style for cloud solutions since it promotes loose coupling, ease of scaling and integration with third-party services, which are fundamental stipulations of BaaS platforms. The methodology adopted in designing a microservice-based BaaS platform was formed in accordance with an Application Programming Interface (API)-first approach, which strives to design a suitable, representative API of the platform. To the best of authors’ knowledge, this paper proposes the lowest-level design of a BaaS platform so far, describing the entity relations, integration patterns, and communication styles. Ultimately, the proposed design was implemented and tested for its functional requirements. In that regard, specific test cases that mirror the actual workflow of the BaaS platform were constructed.

In the field of machine learning, the problem of class imbalance considerably impairs the performance of classification algorithms. Various techniques have been proposed that seek to mitigate classifier bias with respect to the majority class, with simple oversampling approaches being one of the most effective. Their main representative is the well-known SMOTE algorithm, which introduces a synthetic instances creation mechanism as an interpolation procedure between minority instances. To date, an abundance of SMOTE-based extensions that intend to improve the original algorithm has been proposed. This paper aims to compare the performance of several such extensions. In addition to comparing the overall performance, the impact of the selected oversamplers on the per-class performance is also evaluated. Finally, this paper tries to interpret the obtained performance results with respect to the internal procedures of oversampling algorithms. Some interesting findings have been made in this regard.

Continuous optimisation in different technical systems is an ever ongoing, but in many cases a challenging task. Evolutionary algorithms and, in particular, differential evolution (DE) have shown their potential as effective means of optimisation in numerous such instances, especially when the systems are “black boxes”. Yet, in an endeavour to increase the performance of these algorithms even further, various modifications and extensions are steadily introduced. This paper is just another step in the pursuit of gaining additional performance. Correspondingly, a DE extension, one that might benefit not only the canonical algorithm, but also already modified or extended varieties is proposed to this end. The proposal features a straightforward incorporation into the algorithm and is based on the preselection, via a simple surrogate model, of multiple candidate centroids of the entire population that are infused with self-adaptive normal noise. The soundness and viability of its design have been shown experimentally by incorporating it into different DE variants and evaluating the performance gains on a diverse set of challenging benchmark functions.

Wrapper-based feature selection plays a pivotal role in data mining, operating to reduce dimensionality and identify relevant features within datasets. Given the computationally demanding nature of the wrapper’s search for intricate feature relationships, numerous bio-inspired algorithms have been employed to facilitate the process. Notably, the genetic algorithm stands out due to its representation of solutions as binary strings. The literature presents a multitude of genetic algorithm-based wrappers, predominantly employing the n -point crossover operator, where n conventionally takes values of 1 or 2. This study explores the impact of varying the parameter n in the n -point crossover operator on the efficacy of the wrapper’s search. The performed analysis underscores that no single parameter value prevails, motivating the need for dynamic adjustment during the search. Consequently, several elegant strategies for this purpose are proposed and meticulously evaluated, leveraging a comprehensive examination of the genetic algorithm’s convergence behavior. These strategies are experimentally compared with established crossover operators from the literature, leading to the identification of noteworthy discoveries. The empirical findings present a valuable resource for researchers and practitioners alike, poised to enhance feature selection processes within data mining applications. Keywords: bio-inspired optimization; classification; feature selection; genetic algorithm; n -point crossover; wrappers