A LIGHTWEIGHT AI-BASED FRAMEWORK FOR MSME NETWORK PERFORMANCE OPTIMIZATION

Abstract

Small businesses are now required to be efficient in all aspects of their operations, including hiring staff members, spending less money, and cutting back on extra cost. This also applies to network infrastructure. The majority of MSMEs utilize a network that is neither configured nor dynamic and is rarely monitored after installation since they cannot afford to hire a dedicated IT specialist. In most situations, these arrangements work fairly well. As a result, push them during times of high usage, peak traffic, sudden spikes in demand for cloud-based apps, and all open gaps: Uploads become slower, calls begin to decease, and pages begin to index. In this paper, we describe an AI-driven lightweight framework that uses a controlled simulation using NS-3 to address this problem. In order to model and learn the network using supervised learning models, this method employs two network simulations: one for normal MSME operation and one for generating the necessary congestion. A Random Forest (RF) classifier that can differentiate between normal and congested traffic and only launch a set of targeted and limited actions when there is congestion is trained using the flow level KPIs that are extracted from the simulations. Initial stress test shows good results: end-to-end latency decreased by approximately 37% and there was no change to throughput. The main drawback of this study is that everything has been tested and simulated, the next step is to replicate in real life. The goal of this study is to apply artificial intelligence (AI) to optimize MSME network infrastructure.