Serialise to a plain dict for JSON responses.

Validate input data for security threats. Args: data: Input data to validate field_name: Name of the field b

Add console handler with colored output.

Add file handlers for different log levels.

Analyze how diseases are distributed based on symptom count and weight magnitude. Since the model uses hardcoded weights rather than

Rank diseases by diagnostic complexity based on symptom overlap with other diseases.

Analyze which symptoms appear across how many diseases and their average weight.

Create a summary of the symptom overlap between diseases. (Full matrix can be large, so we return top overlapping pairs.)

Create a histogram of symptom counts.

An overall 0-100 balance score for the model. Higher = more balanced.

Compute overall bias indicators across the model.

Core Grad-CAM computation. 1. Build a sub-model that outputs (conv_activations, final_predictions). 2. Forward-pass with GradientTape to

Score-CAM heatmap computation for a TFLite interpreter. Parameters ---------- interpreter : Allocated TFLite Interpreter.

Add profile columns to existing SQLite databases created before this model update.

Walk the model layers in reverse and return the name of the last Conv2D (or Depthwise/Separable Conv) layer. For ResNet50 this is reliab

Find a tensor index that has live runtime data after invoke(). We do a dummy forward pass first, then check which 4-D spatial tensors act

Identify diseases that may be underrepresented due to: - Very few symptoms (below MIN_SYMPTOMS_THRESHOLD) - All low weights

Identify symptoms that appear in only one disease or have consistently low weights.

High-level summary statistics.

Auto-generate display names from symptom keys

Generate realistic vitals based on disease and age.

Get all symptoms from all diseases except the given one.

Get unique identifier for the request. The identifier is based solely on the client IP address. Using User-Agent as part of

Load a fresh TFLite interpreter for Score-CAM. XNNPACK is explicitly disabled because it optimises away intermediate tensor buffers, mak

Compute the Gini coefficient (0 = perfect equality, 1 = total inequality).

Global BMI impact on disease risk.

Simulate predictions with random symptom subsets to estimate per-disease performance characteristics (since we don't have a test data

Forward pass that also captures the intermediate feature tensor. Returns ------- feature_map : float32 array of shape (H_feat, W_fea

Return True only if the leading bytes identify a supported image format.

Add posterior probability column to dataframe

Log after request. Args: response: Flask response object Returns: Response object

Process user's chronological prediction history, calculating: 1. progression timeline with timestamps 2. vitals trend progres

Calculate posterior probability using Bayes' Theorem.

Log before request.

Build a concise explanation summary for the diagnosis.

Check if request is within rate limit. Args: endpoint_type: Type of endpoint (default, prediction, ml_analysis, report)

Clean and validate data

Generate a chat response using Gemini API, restricted to medical/health domain. Args: message: The user's query history: Op

Generate AI-powered recommendations using Gemini API based on disease probability results. Args: disease_name: Name of the disease

Generate Text-to-Speech audio from the provided text using gTTS. Args: text (str): The text to be converted to speech. langu

Get or create the singleton BiasAnalyzer instance.

Get current active model version.

Fetch dashboard data specific to a patient (user). Args: user_id: The ID of the currently logged-in user Returns: dict:

Fetch real dashboard data from the PredictionHistory table. Returns: dict: Dashboard metrics and risk distribution data from databas

Return the strongest positive and negative feature impacts.

Get all version history.

Initialize error handlers for Flask app. Args: app: Flask application instance

Log error with details. Args: error_type: Type of error message: Error message **kwargs: Additio

Log disease prediction. Args: disease: Disease name symptoms: List of symptoms probability: Pred

Log security event. Args: event_type: Type of security event message: Event message severity: Ev

Read CSV data for batch processing

Execute a complete bias and coverage analysis. Returns a comprehensive result dictionary suitable for JSON serialization and dashboar

Save results to CSV