Machine Learning Chaos - Data Chaos

Often I keep hearing below responses for building ML use cases. There are multiple missing dimensions for a judgement. Some of my observations from the discussions

• Data Unavailable - You cannot map a standard use cases and expect those standard features be part of data. ML use cases today go with "I need all this data", rather we should look at "What model I can build with available data"
• Data Insufficient - Data gets archived and deleted in most transactional systems. "Building a model with currently available data and improving it periodically is more important than waiting for 5 year data"
• Model Accuracy - Don't compare ML with other software metrics, ML is learning from data. Garbage In Garbage out. Software is code for functionality
• Picking Right Use-case - Before finding the right use cases, we need to understand the collected data and features. Gap between use cases and understanding of available data will not get you right use cases to solve
• Data Pipeline - Building first ML use case involves setting up data pipeline for collecting newer features, changes to transactional system. This needs good analysis of gap between features available and good to have features with the pipeline setup for adding newer feature variables

#MachineLearning #Data #Business #AI. Navigate through the chaos of Data and Features to build your ML Models.

Happy Learning!!!