Stock price forecasting is highly challenging due to the non-linearity, dynamism, and inherent noise in stock prices, making accurate predictions difficult for investors and traders. This study focuses on denoising stock market time series data by combining two filtering techniques, unlike conventional methods that typically use only one. We analyze historical closing prices from 894 stocks across the US, India, China, and the UK, using filters like Fourier and wavelet transforms. A deep autoencoder based on convolutional neural networks was trained on a dataset of 89400 samples and tested against various techniques, including simple moving average, exponential moving average, and Kalman filter. Our findings highlight the potential of integrating multiple filters, showing improved denoising performance of at least 10%, especially for smaller time steps as in ‘AKAM’ stock. This research is significant for investors and traders looking to automate stock market forecasting.