Historical manuscripts dating remains a significant challenge in the fields of document analysis and classification, and digital preservation. This study aims at investigating a hybrid machine learning model for ancient manuscripts dating. This later integrates the Scale-Invariant Feature Transform (SIFT) with advanced Convolutional Neural Networks (CNNs). To allow the encapsulation of manuscripts intricate characteristics, the model leverages SIFT to extract and encode local features that are afterwards organized into a comprehensive visual codebook. A proposed custom-designed CNN architecture exploits these local features to decipher complex visual patterns indicative of specific historical periods. The deep layers of the network learn hierarchical representations that reflect the evolution of script styles, material composition, and textual formats, enabling the system to detect and interpret chronological signatures within the manuscript imagery. The proposed hybrid model is extensively trained and its performances are assessed on the KERTAS dataset to allow for identifying and exploiting subtle visual markers associated with temporal data. The early results demonstrate the model’s substantial potential to streamline the manuscript dating process, offering historians and archivists a robust tool for more efficient historical document analysis including archival materials cataloging.