4. Time Series Data Analysis

In this section, we’ll explore methods for analyzing time series data from Internet of Things (IoT) devices. We’re going to delve into three detailed units, utilizing the Civil IoT Taiwan Data Service Platform for a deeper understanding.

• 4.1. Time Series Data Processing

  We use the sensing data of Civil IoT Taiwan Data Service Platform to guide readers to understand the use of moving average, perform periodic analysis of time series data, and then disassemble the time series data into long-term trends, seasonal changes and residual fluctuations. At the same time, we apply the existing Python language suites to perform change point detection and outlier detection to check the existing Civil IoT Taiwan data, and discuss potential implications of such values detected.

• 4.2. Time Series Data Forecast

  We use the sensing data of the Civil IoT Taiwan Data Service Platform and apply existing Python data science packages (such as scikit-learn, Kats, etc.) to compare the prediction results of different data prediction models. We use graphics to present the data and discuss the significance of the data prediction of the dataset at different time resolutions in the real field, as well as possible derived applications.

• 4.3. Time Series Data Clustering

  We introduce advanced data grouping analysis. We first present two time-series feature extraction methods, Fourier transform and wavelet transform, and briefly explain the similarities and differences between the two transform methods. We introduce two different time series comparison methods, Euclidean distance and dynamic time warping (DTW), and apply existing clustering algorithms accordingly. We then discuss data clustering with different temporal resolutions, as well as their representation and potential applications in the real world.