| Trial Application and Expected Outcome |
- Stage 1 Data Discovery, Mining and Planning: This stage first discovers the data structure and database characteristics. Data mining procedures will be done to identify correlations to prepare the training data for AI model, including the K-means clustering, kNN classification, SVM clustering etc. to formulate a solid modelling plan.
- Stage 2 RNN Model Formulation and Training: After completing the model architecture and data pipelines, training data will be fed into the model for training. Such training lets the RNN learns the correlation between multi-dimensional data, thus producing a comprehensive prediction on spare parts demand, availability, maintenance schedule etc.
- Stage 3 Applying NLP to summarize user remarks: By using NLP techniques such as Word Segmentation, Algorithmic Tokenization and Unsupervised Keyword Extraction, meaningful keywords will be extracted from all the data, thus to summarize and categorize the data into useful clusters.
- Stage 4 Model Fitting and Fine-Tuning: After the model is developed, it will be fed with unseen real data for testing. Evaluation will be conducted to see the performance of the model, parameters such as accuracy will be assessed. The model can be further fine-tuned with the clustered user remarks.
- Stage 5 Deployment: Due to the limitation on using local servers only, the deployment will be physically done in EMSD’s local connection of the server. The deployment includes code transfer and full deployment testing.
|
