Imputation of missing data in time series is a critical challenge that significantly impacts the accuracy and reliability of predictive models across various fields. This study investigates the effectiveness of a self-defined lightweight 1-dimensional Convolutional Neural Network (CNN) model compared to traditional Machine Learning (ML) methods for addressing this issue. By transforming univariate time series data into a multivariate format, we exploit the inherent temporal relationships within the dataset to enhance the imputation of missing values. We evaluate the performance of the CNN model against several ML algorithms, including Random Forest, Support Vector Regression, Extra Trees, and K-Nearest Neighbors, using datasets from Vu Quang and Hanoi. Experimental results consistently demonstrate that the CNN model outperforms traditional ML methods across various gap sizes, achieving lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), as well as higher Similarity, Nash-Sutcliffe Efficiency (NSE) with actual values. These findings underscore the superiority of CNN as a robust solution for time series data imputation, offering a more effective approach for handling missing data in predictive modeling tasks.