<p>Accurate tool wear prediction optimizes tool replacement strategies, extends tool life, reduces costs, and enhances production efficiency. While recurrent neural networks (RNNs) are widely used for tool wear prediction, they suffer from gradient vanishing, limiting their ability to capture long-term dependencies in sensor data. To address this, we propose an improved temporal convolutional network (ITCNS) for tool wear prediction. First, to enhance the spatial feature extraction capability of temporal convolutional networks (TCNs), we integrate a parallel two-dimensional convolutional module. Second, to mitigate local information loss in TCNs, a parallel structure is designed to extract local temporal features from multiple dimensions. Finally, the extracted features are fused, and the wear value is predicted using a linear layer. Experimental results on the Society for Prognosis and Health Management milling dataset demonstrate that ITCNS achieves high prediction accuracy, with a mean absolute error (<i>MAE</i>) of 4.64, a root mean square error (<i>RMSE</i>) of 6.40, and a coefficient of determination (<i>R</i>²) of 0.966. These improvements facilitate indirect tool monitoring during machining and enhance production efficiency.</p>