Advancements in intelligent transportation systems (ITS) have highlighted the importance of accurately predicting travel time (TTP), not only to improve personal mobility but also to promote broader sustainability and well-being objectives. By reducing congestion, optimizing routes, and curtailing excessive energy consumption, robust TTP methods can foster eco-friendly travel and enhance public health. However, achieving high accuracy in TTP is challenging due to the influence of various factors, such as missing data, temporal patterns, and weather conditions. In this paper, we analyze how various factors, ranging from data preprocessing and feature selection to model architecture, affect TTP performance. Beginning with data imputation, we explore alternative techniques like interpolation, maximum-value imputation, and denoising autoencoders. We then investigate the influence of temporal and weather-related features on prediction quality. Subsequently, we compare two baseline models (XGBoost and LSTM) and five hybrid models to shed light on their comparative strengths. Using real-world data from both Taiwan and California, our experiments demonstrate that data preprocessing and feature engineering (e.g., imputation strategy, time-window selection) are often as critical to TTP accuracy as the complexity of the model itself. Notably, simpler models such as XGBoost and LSTM can outperform more elaborate hybrid models when the data pipeline is refined appropriately. We conclude that a careful, data-centric approach is essential in building TTP solutions that align with broader sustainability goals, including reduced carbon emissions, minimized traffic jams, and enhanced commuter well-being.