Bima is one of the regions in West Nusa Tenggara Province. The Bima script is a cultural heritage used as a means of communication by the Bima community in the past. The decline in the use of the Bima script threatens cultural heritage. The government has addressed this issue by providing training to teachers to teach it in schools, but this has still been insufficient due to the limited number of teachers participating in the training. Therefore, one efficient method to assist with this issue is by leveraging modern technology, particularly through machine learning for handwriting recognition. This study aims to find the best CNN model for recognizing the Bima script with diacritics to help preserve Bima's cultural heritage through handwriting recognition. The CNN model is combined with hyperparameter tuning, and then testing is conducted in four different scenarios to evaluate the performance of each model architecture and hyperparameter variation to find the best combination. The dataset used is sourced from the Kaggle platform, and augmentation is performed to increase the total number of images to 6,750, with each image containing 75 images in 90 different classes. In this study, testing is done by dividing the dataset into training and testing sets in an 80:20 ratio. The test results show high performance, achieving an accuracy of 98.00%, precision of 98.19%, recall of 98.00%, and f1-score of 98.00% in scenario 4.