Data Mining and Fusion Framework for In-Home Monitoring Applications.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Sensor Data Fusion (SDF) is crucial for integrating data from various sources.
• Handling heterogeneous and complex datasets presents a significant challenge in SDF.
• Existing SDF methods often struggle with diverse data formats.

Purpose of the Study:

• To propose a novel Sensor Data Fusion framework capable of handling both homogeneous and heterogeneous datasets.
• To compare the efficacy of data mining software packages for sensor data fusion.
• To develop a data fusion framework specifically tailored for in-home applications.

Main Methods:

• Utilized both homogeneous and heterogeneous datasets, including privacy-friendly binary images and thermal/Radar sensing data.
• Compared data mining software packages: RapidMiner Studio, Anaconda, Weka, and Orange.
• Implemented machine learning models: Naïve Bayes, Decision Tree, Neural Network, Random Forest, SGD, SVM, and CN2 Induction.

Main Results:

• The proposed SDF framework achieved 84.7% average Classification Accuracy on homogeneous datasets.
• Achieved 95.7% average Classification Accuracy on heterogeneous datasets.
• Cross-validation yielded high performance: 94.4% Classification Accuracy, 95.7% Precision, and 96.4% Recall.

Conclusions:

• The novel SDF framework effectively fuses homogeneous and heterogeneous data, outperforming existing methods.
• The framework offers significant cost and time savings in data labelling, preparation, and feature extraction.
• The proposed approach is suitable for in-home applications, improving data integration efficiency.