Deep Learning–driven Decision Support System for characterization of probability distribution tails
Abstract
A novel, fully automated Deep Learning-based DSS that classifies an input dataset into ten distinct probability distributions viz. Lognormal (Sub-exponential case with α>1), Lognormal (Hyper-exponential case with α<1), Weibull (Sub-exponential case with α<1), Weibull (Hyper-exponential case with α>1), Gamma (Sub-exponential case with α<1), Gamma (Hyper-exponential case with α>1), Pareto, Exponential, Normal and Log-Pearson III. The framework integrates seven task-specific Deep Neural Network models trained on concentration profiles, concentration-adjusted expected shortfall, Zenga curves, Hill ratio plots, and Zipf plots.
Full text
Description: The repository contains seven trained Keras models and their corresponding scalers (.pkl files). The notebook named DSS_complete_code.ipynb is a ready-to-use Google Colab script designed for automated characterization of probability distribution tails. How to Use the Decision Support System (DSS): 1. Upload the file DSS_complete_code.ipynb to Google Colab. 2. Load all seven trained models and their corresponding scalers. 3. Load your input dataset and convert it to a NumPy array. 4. Call the function DSS_classification_model(), passing the input dataset as an argument. 5. The function will return the corresponding best-fit probability distribution for the dataset. Notes / Recommendations: 1. Ensure that all model (.h5) and scaler (.pkl) files are in the same working directory as the notebook. 2. The notebook includes all necessary dependencies; no external installation is required apart from standard TensorFlow/Keras and NumPy libraries.