Understanding the influence of physical activity on glycemic control is essential for effective management of Type 1 diabetes mellitus. However, its consideration in clinical decision-making and automated glucose regulatory devices are often insufficient. This study aims to address this gap by utilizing real patient data to assess the effects of physical activity. Long short-term memory (LSTM) models, known for their ability to capture time dependencies, can be used to analyse physical activity patterns and their impact on glycaemic status. Additionally, we investigate various parameters affecting LSTM performance, such as input sequence length, number of hidden units, and learning rates. Through this analysis, we aim to provide insights into how these parameters affect LSTM performance in accurately predicting physical activity affects on glycemic state. By exploring the impact of different LSTM parameters, we seek to enhance our understanding of personalized diabetes management strategies tailored to individual patient needs.