Built to excel in IoT, machine vibration monitoring, and wearable devices!
Built to excel in IoT, machine vibration monitoring, and wearable devices!
Concatenated Decision Paths (CDP) is a lightweight, efficient time series classification model that uses shapelet-based decision trees to extract critical patterns, offering robust, accurate classification with minimal computational requirements, making it ideal for diverse datasets and resource-constrained environments like IoT, machine vibration and wearables.
CDP has demonstrated high accuracy across various benchmark datasets from the UCR archive, often outperforming traditional methods in both training time and classification accuracy. For example, on the 'Swedish Leaf' dataset, CDP achieved an accuracy of 92.7% with a training time of 16.3 seconds on Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz.
The CDP method consists of several key steps:
1. Subsets Extraction: All possible subsets of 2 (rarely 3, or 4 ) class labels are extracted from the original dataset. Each subset is used to create a decision tree.
2. Decision Trees Training: For each subset, a decision tree is trained using a particle-swarm-optimization (PSO) environment to discover the most discriminative shapelets for two classes. The decision trees consists of two leaves, representing the subset.
3. Path Collection: During the classification process, each time series traverses through the decision trees, generating a decision paths. Every decision tree has shapelet associate with that tree. If the distance between that shapelet and incoming time series (TS) is less than preliminary found split distance, then TS is classifies to belong to the ‘left’ leave, otherwise it goes to the right leave. In every such case decision path adds either ‘L’ or ‘R’.
4. Pattern Formation: The decision paths are concatenated to form a decision pattern unique to each class. This pattern is then used for classification. For example if 10 decision trees are present then incoming time series (TS) will get ‘L’ or ‘R’ from every such tree. After all, concatenating these path, TS gets specific pattern that looks like: ‘LLLRLLRLLL’.
5. Patterns comparison: Every TS from the training dataset has such pattern. Comparing pattern of the incoming time series with ones collected during training process, brings final decision which is the class of the closest TS from train dataset.
Acquire expertly designed, space- and speed-optimized source code for the C++ implementation of CDP model. The package includes source files along with project configurations for Windows and Linux. The code is versatile and can run efficiently on devices ranging from compact microcontrollers, such as the Raspberry Pi Pico, to high-performance computing systems.
All purchases are securely processed through PayPal. Please review our Privacy Policy (outlined below) prior to completing your transaction.
Upon successful purchase, an access link to the project will be sent to the email address provided during checkout. If you do not see the email in your inbox within a few minutes, please check your junk or spam folder, as emails can sometimes be filtered there. For any issues, feel free to contact our support team.
If you prefer not to use PayPal and would like to pay via a bank transfer, please contact us through the 'Get in Touch' form. Include a clear description of your payment preferences and any specific details required to process the transaction. Our team will provide you with the necessary bank account information and further instructions.
We appreciate your support and thank you for your purchase!
Send an enquiry about CDP or customized solution, based on CDP algorithm.
We use cookies to analyze website traffic and optimize your website experience. By accepting our use of cookies, your data will be aggregated with all other user data.