Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ED-Filter: Dynamic Feature Filtering for Eating Disorder Classification

Mehdi Naseriparsa, Suku Sukunesan, Zhen Cai, Osama Alfarraj, Amr Tolba, Saba Fathi Rabooki, Feng Xia

TL;DR

ED-Filter addresses high-dimensional, dynamic Twitter data for eating-disorder classification by combining an informed branch-and-bound feature search with a hybrid greedy-deep learning strategy. It ranks features using Information Gain and employs an admissible upper bound $\overline{\theta}(F,Y)$ in a branch-and-bound framework, while a neural cardinality detector guides subset size; a greedy sub-solution and a deep-learning–assisted hybrid further accelerate search. Experiments on 11,620 Pro-ED Twitter accounts show that ED-Filter achieves competitive or superior classification accuracy while reducing feature counts and maintaining tractability in streaming contexts, compared with filter and wrapper baselines. The approach enables a practical, adaptive ED-detection pipeline for social media and points to extensions to other platforms and multi-modal signals.

Abstract

Eating disorders (ED) are critical psychiatric problems that have alarmed the mental health community. Mental health professionals are increasingly recognizing the utility of data derived from social media platforms such as Twitter. However, high dimensionality and extensive feature sets of Twitter data present remarkable challenges for ED classification. To overcome these hurdles, we introduce a novel method, an informed branch and bound search technique known as ED-Filter. This strategy significantly improves the drawbacks of conventional feature selection algorithms such as filters and wrappers. ED-Filter iteratively identifies an optimal set of promising features that maximize the eating disorder classification accuracy. In order to adapt to the dynamic nature of Twitter ED data, we enhance the ED-Filter with a hybrid greedy-based deep learning algorithm. This algorithm swiftly identifies sub-optimal features to accommodate the ever-evolving data landscape. Experimental results on Twitter eating disorder data affirm the effectiveness and efficiency of ED-Filter. The method demonstrates significant improvements in classification accuracy and proves its value in eating disorder detection on social media platforms.

ED-Filter: Dynamic Feature Filtering for Eating Disorder Classification

TL;DR

ED-Filter addresses high-dimensional, dynamic Twitter data for eating-disorder classification by combining an informed branch-and-bound feature search with a hybrid greedy-deep learning strategy. It ranks features using Information Gain and employs an admissible upper bound in a branch-and-bound framework, while a neural cardinality detector guides subset size; a greedy sub-solution and a deep-learning–assisted hybrid further accelerate search. Experiments on 11,620 Pro-ED Twitter accounts show that ED-Filter achieves competitive or superior classification accuracy while reducing feature counts and maintaining tractability in streaming contexts, compared with filter and wrapper baselines. The approach enables a practical, adaptive ED-detection pipeline for social media and points to extensions to other platforms and multi-modal signals.

Abstract

Eating disorders (ED) are critical psychiatric problems that have alarmed the mental health community. Mental health professionals are increasingly recognizing the utility of data derived from social media platforms such as Twitter. However, high dimensionality and extensive feature sets of Twitter data present remarkable challenges for ED classification. To overcome these hurdles, we introduce a novel method, an informed branch and bound search technique known as ED-Filter. This strategy significantly improves the drawbacks of conventional feature selection algorithms such as filters and wrappers. ED-Filter iteratively identifies an optimal set of promising features that maximize the eating disorder classification accuracy. In order to adapt to the dynamic nature of Twitter ED data, we enhance the ED-Filter with a hybrid greedy-based deep learning algorithm. This algorithm swiftly identifies sub-optimal features to accommodate the ever-evolving data landscape. Experimental results on Twitter eating disorder data affirm the effectiveness and efficiency of ED-Filter. The method demonstrates significant improvements in classification accuracy and proves its value in eating disorder detection on social media platforms.
Paper Structure (18 sections, 2 theorems, 7 equations, 6 figures, 3 tables, 2 algorithms)

This paper contains 18 sections, 2 theorems, 7 equations, 6 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Eating Disorders
  4. Feature Selection
  5. Problem Definition
  6. The Proposed ED-Filter Method
  7. Ranking the Features
  8. Iterative Feature Reduction
  9. Feature Cardinality Detector
  10. Branch & Bound Search Algorithm
  11. The Greedy-based Sub-optimal Solution
  12. The Hybrid Solution
  13. Experimental Analysis
  14. Data Collection, Cleaning, and Processing
  15. Effectiveness Assessment
  16. ...and 3 more sections

Key Result

Lemma 1

The upper bound value of the classification accuracy for a set of features $F=\{f_1,f_2,...,f_n\}$ based on its mutual information is computed as follows:

Figures (6)

  • Figure 1: The architecture of ED-Filter.
  • Figure 2: The steps in ED-Filter.
  • Figure 3: Effectiveness Analysis, figure (a), (b), (c), (d) show the effectiveness of greedy and hybird methods, (e) and (f) demonstrate the effectiveness difference between greedy and hybird methods.
  • Figure 4: ED-Filter versus Other Methods.
  • Figure 5: Scalability of hybrid greedy-deep learning method.
  • ...and 1 more figures

Theorems & Definitions (9)

  • Definition 1
  • Definition 2
  • Definition 3
  • Definition 4
  • Lemma 1
  • proof
  • Lemma 2
  • proof
  • Definition 5