The success of any machine learning model heavily relies on the type of input data it receives. The more reliable character data have greater the variety, the better the model can be trained, resulting in more accurate outcomes. However, acquiring a large volume of diverse data can be quite challenging in certain cases. One such case is the handwritten character reputation for the Kaithi script. This paper presents a comprehensive approach to recognizing handwritten characters in the Kaithi script, utilizing a character known as the Kaithi Dataset. The Kaithi dataset is comprised of character images sourced from both historical documents and manually created samples. The dataset is organized into training and testing subsets, facilitating a structured approach for model evaluation. The images are presented in RGB format and capture a diverse range of character styles. To improve model performance and mitigate overfitting, image variations were introduced as part of the preprocessing steps. This dataset was employed to assess the influence of these variations on recognition accuracy, highlighting their significance in enhancing the effectiveness of handwritten character recognition tasks.