Anomaly / Novelty Detection Analysis

Back to Services
Upload data
Identify columns
Check data
Review and start
Start a guided analysis

Anomaly / Novelty Detection Analysis

We are loading the service guidance for this workflow.

You will identify the key columns after uploading your data.
Need help preparing your data?

Preparing workflow guidance...

We are loading the dataset structure and key columns for this workflow.

Guidance will appear here once this service loads.

                        

                    

                


                
    
    
    

    

        

            
Dataset Setup

            
Start with how your data is already split.

            
Choose this first. AutoML will turn on only the upload fields and column choices that match that setup.

        

    

    

        

            
            
            
At the default setting, AutoML creates separate training, validation, and locked-test portions from this file.

        

        

            

                
Current setup

                
One file; let AutoML split it

                

                    Default: approximately 60% trains models, 20% checks model choices, and 20% is held back for final reporting.
                

            

        

    

    

        

            

                

                    
2. Upload the matching file or files

                    

                        Upload one dataset. Validation and test file uploads stay inactive for this setup.
                    

                    

                        Choose Separate train, validation, and/or test files above to open dedicated upload cards for each role.
                    

                

            

        

        

            

                

                    

                        

                            
One-file setup

                            
                        

                        Required
                    

                    
                    
Required. AutoML reads its columns for mapping, optional split-column choice, and optional sampling weights.

                

            

            

                

                    

                        

                            
Fits the models

                            
                        

                        Required
                    

                    
                    
Required. Used to learn preprocessing and fit the selected models. Validation and test files must use matching columns.

                

            

            

                

                    

                        

                            
Checks choices

                            
                        

                        Optional
                    

                    
                    
Optional. Used for model checking, tuning, calibration, or threshold choice when supported. It is not the final score.

                

            

            

                

                    

                        

                            
Final score

                            
                        

                        Optional
                    

                    
                    
Optional. When supplied, this is the final holdout used by the main Model Performance table and reports.

                

            

        

    

    

        
3. Optional columns

        
These lists populate after the dataset or training file is read.

        

            

                
                
                
Choose the column with values such as train, validation, valid, test, or holdout. AutoML excludes it from model features.

            

            

                
                
                
Optional. Use when rows represent a weighted sample of the population.

            

        

    




    
    

    
    

      

        
          
        
Optional known-anomaly label for evaluation. Leave blank for unsupervised detection.

      


      
      

        
        
        
          
        
        
If a target column is provided, choose the anomaly or event label used for metrics.

      

    


    


                

                    

    

        

            
Check your data

            
Find problems before training starts.

            
Check the file for missing information, unsuitable columns, and other issues that may affect the results.

        

        

            
            
No charge. No training starts.

        

    


    

        

            

                
Data check

                
What needs your attention

                
Resolve blocking issues, review warnings, then choose models and start training.

            

            No preview yet
        

        

    



                


                

                

                    

                        

                            
Step 4

                            
Choose what to compare, then start

                            
Your recommended setup appears after the data check is complete.

                        

                    


                    

                        

                            

                                
Recommended setup

                                
Start with the balanced comparison.

                                
AutoML will compare a practical mix of suitable models. Change it only when you have a specific reason.

                            

                            
                        

                        
                        

                            
Choose a comparison approach

                            
Balanced comparison is selected by default. Choose another approach only when its time and cost trade-off fits your goal.

                            

                                
                                
                                
                                
                                
                            

                        

                    


                    


                    

                        

                            

                                
Before you start

                                
Review the cost and expected outputs

                                
Your estimate appears after you choose a setup.

                            

                            

                                Analysis type loading
                                Higher-cost models
                            

                        

                        
                        

                            See how this estimate was calculated
                            

                                
Type of analysis: Loading...
Cost details will appear here.

                                
Typical cost range: --
The amount held before training will appear here.

                                
Suggested plan: Loading...
Plan guidance will appear here.

                            

                            
Plan limits will appear here once this service loads.

                            
Expected outputs: Metrics summary, report, and prediction-ready files will appear here once this service loads.

                            
We will show the checks this workflow runs after file preview or a readiness preview.

                        

                    


                    

                        
                        Sign in to start training