@article{article_1422119, title={Accurate Indoor Home Location Classification through Sound Analysis: The 1D-ILQP Approach}, journal={Firat University Journal of Experimental and Computational Engineering}, volume={4}, pages={12–29}, year={2025}, DOI={10.62520/fujece.1422119}, author={Abdullahi, Nura and Akbal, Erhan and Dogan, Sengul and Tuncer, Türker and Erman, Umut}, keywords={Home location detection, 1D-ILQP, Neighborhood component analysis, Sound classification, Machine learning}, abstract={Detecting human activities within domestic environments constitutes a fundamental challenge in machine learning. Conventionally, sensors and video cameras served as primary tools for human activity detection. However, our work is oriented towards the innovative objective of ascertaining home locations by analyzing environmental sound signals. Consequently, we compiled a comprehensive sound dataset from eight distinct locations. To enable automatic home location detection using this sound dataset, we employed a lightweight machine learning model designed with a paramount focus on precision and minimal computational overhead. At the core of our approach is the introduction of a local feature generator, referred to as the one-dimensional Improved Local Quadruple Pattern (1D-ILQP). This novel 1D-ILQP plays a central role in the feature extraction process, generating textural features from the acoustic signals. To facilitate the extraction of high-level textural features, we emulated the convolutional neural network (CNN) architecture, applying maximum pooling to decompose signals. The suggested 1D-ILQP extracts textural features from each decomposed frequency band as well as the original signal. Subsequently, we selected the top 100 features using the Neighborhood Component Analysis (NCA) technique. The final step of our model involves classification, wherein we employed a range of classifiers, including decision trees, linear discriminant analysis, quadratic discriminant analysis, Naive Bayes, support vector machines, k-nearest neighbor, bagged trees, and artificial neural networks. We subjected the results to a comprehensive evaluation, and all classifiers achieved classification accuracies exceeding 80%. Notably, the k-nearest neighbor classifier delivered the highest classification accuracy, reaching an impressive 99.75%. Our findings unequivocally demonstrate that the proposed sound classification model, based on the 1D-ILQP, has yielded highly satisfactory results when applied to the home location sound dataset.}, number={1}, publisher={Fırat University}