PaperlayerView exact revision on arXiv

Source-linked AI summary

Right Choice of Classification Algorithms Based on Reinforcement Learning for Prediction of Non-Alcoholic Fatty Liver

Hasan Samadbin, Arman Daliri

arXiv:2609.29181v1cs.AI

TL;DR

The paper addresses the difficulty of choosing among classifiers for PBC prediction when algorithm performance may vary across datasets. It proposes Fourth Degree Learning, a reinforcement-learning method for automatic classifier selection, and reports 98% accuracy for Stage 4 prediction.

  • Problem

    Choosing among many PBC classifiers is time-consuming, and algorithm performance may vary across datasets.

    Source: §1 · p. 1

  • Method

    Fourth Degree Learning uses a voting-inspired reinforcement-learning framework to select a classifier for PBC prediction.

    Source: §3 · p. 2, B. Fourth Degree Learning (FDL) · p. 3

  • Results

    98% accuracy was reported for Stage 4 prediction.

    Source: §5 · p. 4

  • Takeaways & Limitations

    The authors present the framework as potential inspiration for researchers in machine learning and medical studies.

    Source: §5 · p. 4

Abstract

from arXiv · show

There are many complex issues in the world of artificial intelligence. Some of these problems are solved using other artificial intelligence methods, which are called artificial intelligence for artificial intelligence. Finding an appropriate classifier algorithm is a time-consuming task. For this reason, an algorithm that can automatically learn the choice of classification algorithms is very important. Classification algorithms are useful in predicting various diseases. Also, Primary Biliary Cirrhosis is one of the most well-known diseases that have been predicted by classification algorithms. This research's most significant achievement and novelty is the automatic increase in learning through a scoring method of reinforcement learning is called square learning (SL). In this research, an algorithm is presented that learns to automatically select the appropriate classification algorithm to predict Primary Biliary Cirrhosis. In this article, with inspiration from four evaluation metrics in classification algorithms, a new reinforcement learning method by the name of Fourth Degree Learning has been presented. In this research, we increased the performance of the classification algorithms used in this method from 63% of accuracy and achieved 98% accuracy.

1. INTRODUCTION

The paper frames PBC prediction as a setting where classifier choice is difficult because algorithms can perform differently across datasets. It introduces Fourth Degree Learning to automatically select a classifier for PBC prediction.

1. INTRODUCTION

  • Liver diseases cause approximately two million deaths worldwide each year, and PBC is described as a chronic disease for which no specific treatment is available.The introduction also identifies alcoholic and non-alcoholic fatty liver as forms of fatty liver disease.Source: §1 · p. 1
  • Different algorithms can perform differently across datasets, and the many available PBC classifiers make testing them one by one impractical.The paper notes that prediction accuracy may decrease with different datasets.Source: §1 · p. 1
  • The authors introduce Fourth Degree Learning, a reinforcement-learning approach for selecting a classifier to predict PBC using the Mayo Clinic dataset and three classifiers.The classifiers are Decision Tree, Random Forest, and Support Vector Classifier.Source: §1 · p. 2

2. RESEARCH BACKGROUND

The background describes PBC as a progressive liver disease and identifies the Mayo Clinic trial dataset as the study's data source. The dataset covers 424 patients and 20 medical and clinical features.

2. RESEARCH BACKGROUND

  • PBC is described as progressive, with later-stage disease potentially leading to cirrhosis, liver failure, or death.Many people in early stages have no symptoms.Source: §2 · p. 2
  • The study uses the Mayo Clinic trial dataset for primary biliary cirrhosis.The background identifies it as an original PBC dataset.Source: §2 · p. 2
  • The Mayo dataset includes 424 patients seen between 1974 and 1984, with 20 medical and clinical features.It also records deaths after patient enrollment, which the passage says can be used for time-series analysis [30].Source: §2 · p. 2

3. METHODOLOGY OF RIGHT CHOICE OF CLASSIFICATION ALGORITHMS

The methodology presents a three-part framework for early PBC prediction, culminating in reinforcement-learning-based classifier selection and prediction. Its FDL subsection is the central method component.

3. METHODOLOGY OF RIGHT CHOICE OF CLASSIFICATION ALGORITHMS

  • The framework is presented as a reinforcement-learning algorithm for early PBC prediction.It is inspired by a voting system.Source: §3 · p. 2
  • The three-part workflow prepares the dataset, applies voting-inspired reinforcement learning, then selects a classifier and presents a prediction.Source: §3 · p. 2
  • The methodology section divides its account among dataset preparation, the main FDL method, and the framework's final step.Source: §3 · p. 2

A. Data Cleaning

The data-cleaning stage prepares data for machine learning through privacy, null, duplicate, and outlier checks, then addresses invalid data and class balance. The passage names shark smell optimization for balancing.

A. Data Cleaning

  • The first-stage tasks are privacy detection, null detection, duplicate detection, and outlier detection.The framework removes personal characteristics and useless patient IDs, and extracts data figures for intuitive feature analysis.Source: A. Data Cleaning · p. 2
  • Duplicate samples and features with outliers are identified, and duplicative and high-quality examples are removed.Source: A. Data Cleaning · p. 2
  • Invalid and null data are removed, and imbalanced targets are moderated with balancing methods, including shark smell optimization.The passage names Decision Tree, Random Forest, and Support Vector Classifier as the classifiers used.Source: A. Data Cleaning · pp. 2–3

B. Fourth Degree Learning (FDL)

Fourth Degree Learning scores classifiers using precision, recall, F-score, and accuracy, then updates rewards across iterations. Roulette-wheel selection balances ranked performance with a chance for weaker algorithms to be chosen.

B. Fourth Degree Learning (FDL)

  • Each classifier is run initially and scored using precision, recall, F-score, and accuracy to pursue the highest possible score.The score is the Fourth Degree Learning rate.Source: B. Fourth Degree Learning (FDL) · p. 3
  • At each iteration, algorithm rewards are updated and stored in a list, with the highest-scoring algorithm identified as the maximum Fourth Degree Learning rate.The passage also describes accumulating scores across iterations.Source: B. Fourth Degree Learning (FDL) · p. 3
  • Roulette-wheel selection uses the ranked reward list to choose an algorithm, giving weaker algorithms a chance until the iteration-count stopping condition is met.The selected algorithm is executed as the action for that iteration.Source: B. Fourth Degree Learning (FDL) · p. 3

C. Steps of Fourth Degree Learning (FDL)

The framework cycles through data preparation, algorithm execution and selection, scoring, and prediction until its stopping condition is met.

C. Steps of Fourth Degree Learning (FDL)

4. EXPERIMENTAL EVALUATION

Experiments compare balancing and classifier performance across PBC classes, then evaluate the Fourth Degree Learning framework over iterations. The reported overall performance is higher across both classes, with accuracy reaching the 95% range from iteration 80 onward.

4. EXPERIMENTAL EVALUATION

  • Experiments assess Decision Tree, Random Forest, and Support Vector Machine classifiers, with default initial parameters, and report performance for majority and minority classes.The passage identifies Class 1 as Stages 1–3 and Class 2 as Stage 4.Source: §4 · p. 3
  • Overall performance is higher for both PBC classes than for the other methods, although some classifiers lead on individual evaluation metrics.Random Forest has the highest F1-score; the table reports F1-score, precision, and recall for three classifiers over imbalanced data.Source: §4 · p. 4, Table I · p. 4
  • For the Fourth Degree Learning evaluation, classifier accuracy with a balanced dataset is in the 95% range at iteration 80 and above; the highest iteration is best.Source: §4 · p. 4, Table II · p. 4

5. CONCLUSION

The conclusion reports 98% accuracy for Stage 4 PBC prediction using the proposed Fourth Degree Learning classifier-selection method. It also identifies broader datasets, diseases, balancers, classifiers, and online runtime prediction as future directions.

5. CONCLUSION

  • 98% accuracy is reported for Stage 4 PBC prediction by the proposed framework.The framework uses Fourth Degree Learning to select a classifier and is evaluated with balanced and imbalanced data.Source: §5 · p. 4
  • Future work will evaluate the framework on different big-data sets and other diseases, and improve its balancers and classifiers.Source: §5 · p. 4
  • An online version of the framework could be implemented for runtime PBC prediction.Source: §5 · p. 4
Loading 2609.29181v1…