Abstarct:   In this research, a novel method baxsed on convolutional neural networks (CNNs) is proposed for the analysis of spectral induced polarization (SIP) data. This method is developed to enhance both the accuracy and speed of extracting mineral parameters from data. Using a comprehensive simulated dataset, the neural network is trained with various parameter and model response combinations to identify intricate relationships between SIP data and model parameters. These combinations are prepared for model training using various statistical methods such as logarithmic transformations, principal component analysis (both linear and nonlinear), and independent component analysis. Results from applying this method to real data from the Kervian region demonstrate its ability to predict mineral parameters with high accuracy, making it a powerful tool for mineral exploration. The primary advantages of this method include high speed, high accuracy, simplicity, and efficiency in modeling complex relationships between data. In this approach, the generalized effective medium theory model is employed to simulate SIP data, while the CNN is utilized to learn patterns within the data and predict mineral parameters. This approach effectively addresses the challenges inherent in traditional SIP inversion methods, such as complex calculations and nonlinear relationships.