Implementasi Projection Profile dan K-Nearest Neighbors dalam Pengenalan Tulisan Karakter Aksara Lampung

As one of the countries with a rich culture, Indonesia faces considerable challenges to preserve the diversity of its languages and writings. Lampung Province is one of the regions that has its language and script. However, over time, the variety of writings in the form of the Lampung script began to be forgotten because this variety of writings was not often used as a means of daily communication. One effort can be done to digitize the Lampung script, using the character recognition method using the projection profile as the feature extraction method, and K-Nearest Neighbors (KNN) algorithm to predict character shape. This study has two types of data: training data and testing data with 18-character labels. The comparison of the training and testing data used is 75.56% for testing data, and 24.43% for testing data. The features used are features obtained from image extraction measuring 20 x 20 pixels using a projection profile. The types of features obtained are Horizontal Projection Profile (HPP), Vertical Projection Profile (VPP), and Combined Projection Profile (CPP) which are obtained by combining HPP and VPP features. KNN classification is carried out with 10-fold cross-validation, using a base model and an optimized model with GridSearchCV. The best classification results are obtained using the Combined Projection Profile (CPP) feature with an optimized model using GridSearchCV. The evaluation results obtained for the best image data classification with an accuracy of 86.23%, precision of 86.23%, the sensitivity of 86.23%, specificity of 99.19%, F-Measure of 86.23%, and Matthew Correlation Coefficient (MCC) of 85.09%.