State-of-the-art large language models (LLMs) have undergone rigorous training on extensive datasets, resulting in the development of highly complex models with an extensive number of parameters. These advanced models have exhibited a remarkable ability to categorize and generate text tailored to specific tasks effectively, even when they have not been previously exposed to the pertinent data. This cutting-edge technology is commonly referred to as in-context learning (ICL). The primary objective of this research paper is to conduct an in-depth analysis of ICL, encompassing its applications, optimal methodologies, and the critical factors that impact its effectiveness, such as the template, model size, and shot types (i.e., the number of the example used). Our study involves a meticulous evaluation and implementation of the key ICL strategies across two distinct datasets, revealing the best practices for leveraging ICL to achieve high results and enhance overall performance.