Link Search Menu Expand Document
Developer | Sahha

Data Analysis


Table of contents
  1. Terms & Definitions
  2. Current Status
  3. Types of Analysis Sahha provides
    1. Binary Classification
    2. Binary classification use cases:
  4. Multi-class Classification (Coming soon)
    1. Multi-class classification use cases:

Terms & Definitions

  • State: The condition assessed - ie: depression

  • Sub-state: The severity of the condition assessed ie: ‘mild’.

  • Probability: the probability of each state

  • Data source: The data types that have been used to predict the state ie: sleep time, age

  • Specificity: The ability of a model to correctly identify people without the condition - the probability of a negative assessment (AKA true negative rate)

  • Sensitivity: The ability of a model to correctly identify people with the condition - the probability of a positive assessment (AKA true positive rate)

  • Score: The predicted instrument score ie: PHQ9 Score of 16

  • Sub-state probabilities: the probabilities for each sub state ie: 0.9 [high probability of no depression], 0.3 [low probability of mild depression]…

  • Score confidence interval: the standard deviation ie: 13, 19 2 standard deviations from a score of 16

  • Score error: the score error margin ie: +-3 from score of 16


Current Status

Model status will be regularly updated here: Developer Blog


Types of Analysis Sahha provides

Currently Sahha provides two levels of analysis, a “simple” version and a more data rich version. More detailed and immediately available analysis types will be available soon.

Binary Classification

A “simple” analysis providing a binary output of either depressed or not depressed as well as a probability score indicating the likelihood the analysis is correct.

Binary Classification output:

{
   "inferences":[
      {
         "createdAt":"2022-05-20T00:30:00+00:00",
         "modelName":"automl_toolkit_randomForest",
         "predictionState":"not_depressed",  //binary classifiction either depressed or not depressed
         "predictionSubState":"",
         "predictionRange":-1,
         "predictionConfidence":0.8, //model probability - 0.75 means 75% confident in the prediction
         "dataSource":[
            "sleep",
            "screenTime"
         ], //data types used in model
         "dataSourceSummary":[
            {
               "type":"sleep",
               "amount":"0"
            },
            {
               "type":"screenTime",
               "amount":"0"
            }
         ]
      }
   ]
}

Binary classification use cases:

  • Provide a simple indication of mental state of a user
  • Use to enrich data for better feature delivery

Multi-class Classification (Coming soon)

A more feature rich classification output providing more detail and sub-state analysis.

Multi class classification output:

{
	probability: 0.8, //model specificity
	state: "depressed", //binary classifiction either depressed or not depressed
	substate: "moderate", //multi class classifiction either {none, mild, moderate, severe}
	substate_probabilities: [0.9, 0.1, 0, 0], //probabilities for each state
	score: 16, //phq9 score predicted
	score_confidence_interval: (13, 19), //2 stddevs
	score_error: 3 // 2stddevs
	data_souce: ["screentime", "sleeptime", "age", "gender"] //data types used in model
}

Multi-class classification use cases:

  • Provide a detailed indication of mental state of a user
  • Use to enrich data for better feature delivery
  • Use to deliver specific interventions with more confidence