Classification of House Categories Using Convolutional Neural Networks (CNN)

Abstract

Image classification of house categories, such as condominium, detached house, shophouse and townhouse, is still a major pain point and crucial burden for websites’ backend works. Because, this process is still using manual classification, which are low productivities and often found errors. The objective of this research is to develop an appropriate model of the “Convolutional Neural Networks (CNN)” for house categories classification. The results from 4 models processing revealed that the Based model return the highest overall accuracy value of 75.00 percent. The second, third and fourth overall accuracy values were MobileNet, of 73.24 ResNet50 of 72.78, and VGG-19 of 25.00 percent respectively. The MobileNet model had the highest value of Precision. While the Based model also had the highest values of Recall and F1-score, comparing to other three pre-trained models. In conclusion, the Based model was the most appropriate model to this research. However, the three models, including Based model, ResNet50 and MobileNet models, had small different accuracy values, which can be used for house categories classification.