Predictive analytics technique based on hybrid sampling to manage unbalanced data in smart cities

Area of Science:

• Smart City Technologies
• Data Science
• Artificial Intelligence

Background:

• Smart cities rely on Artificial Intelligence (AI) for autonomous decision-making, requiring extensive data from the physical world.
• Internet of Things (IoT) sensor devices collect environmental data for predictive analytics, but this data is often imbalanced.
• Imbalanced data, if not pre-processed, can lead to significant errors in AI-driven decisions.

Purpose of the Study:

• To propose a novel predictive analytical technique for managing imbalanced data in smart city IoT environments.
• To develop a robust pipeline integrating data pre-processing and machine learning for accurate predictions.
• To enhance the decision-making capabilities of smart cities by addressing data imbalance challenges.

Main Methods:

• A pipeline was designed incorporating Principal Component Analysis (PCA), a hybrid sampling method (SMOTE+ENN), and a Machine Learning (ML) prediction model.
• SMOTE+ENN was employed to transform imbalanced datasets into a balanced state.
• ML algorithms were utilized for clustering and making predictions on the processed dataset, using a large Smart City IoT dataset (405,184 records).

Main Results:

• The proposed technique accurately predicts human presence near IoT devices.
• Evaluation metrics including accuracy, precision, recall, F1-score, and Area Under Curve (AUC) demonstrated strong performance.
• For cluster 0, accuracy was 0.79, precision 1.0, recall 0.79, F1-score 0.87, and AUC 0.88. For cluster 1, accuracy was 0.86, precision 0.99, recall 0.86, F1-score 0.92, and AUC 0.92.

Conclusions:

• The developed technique effectively manages imbalanced data for predictive analytics in smart cities.
• The proposed method shows promise for improving decision-making in various real-world applications.
• The integration of PCA, SMOTE+ENN, and ML offers a robust solution for smart city IoT data challenges.