Enhanced Recognition of English Handwriting Styles Using a Hybrid CNN-Random Forest Model

Most organizations worldwide focus on handwriting recognition as it is vital in several assignments like the archive, education, and security. This is a research paper that puts forward handwriting type recognition employing CNNs and Random Forest classifiers. The study focuses on the recognition of four distinct English handwriting styles: There are four main styles which are; Italics, Spencerian Script, Palmer method, and Zaner-Bloser method. The preprocessing and data enhancement processes were carefully performed so that the model would have 1,000 samples for each of the two styles of handwriting. The hybrid model thus utilizes the CNN feature of extracting discriminative features from the handwriting samples, and the Random Forest classifier. The performance of the hybrid model was further assessed using measures such as; accuracy, logarithmic loss, MCC, and AUCROC. The results demonstrated that the hybrid CNN-Random Forest model outperformed standalone CNN and Random Forest models, achieving an accuracy of 91.2%, a logarithmic loss of 0.35, an MCC of 0.85, and an AUC-ROC of 0.92. Further validation of the said model and its efficiency was done by comparing epoch-wise as well as the batch size-wise results. Furthermore, the hybrid model was done in comparison with the other category of recognition known as handwriting recognition, and shown that the hybrid model has a greater performance.