An Intelligent Detection and Analyzing Mobile Platform for Pollen Allergies Prediction using Artificial Intelligence and Data Science

Chengyu Liu, Julian Avellaneda2023 · DOI

TL;DR

This paper introduces "Allergy App," a mobile platform built with Flutter that leverages machine learning to provide personalized pollen allergy risk assessments. By combining environmental factors with individual health data (like asthma history and smoking habits), the app generates a tailored risk score and actionable medical advice to help users avoid severe allergic reactions.

What Problem Does This Paper Address?

While standard weather apps often report pollen counts or air quality indices (like P.M. 2.5), they rarely translate that data into personal risk. For a person with a history of asthma, a "moderate" pollen day might be life-threatening, whereas a healthy individual might only experience a minor sneeze.

The authors identify a gap in "tailored" health informatics: existing tools provide general environmental data but fail to account for the user's specific physiological vulnerabilities. The goal of this research is to move from general broadcasting to personalized prediction, specifically aiming to prevent the onset of asthma attacks triggered by seasonal allergens.

Methods

The project is split into a cross-platform mobile frontend and a machine learning backend:

  • Frontend Development: The app was developed using the Flutter framework, chosen for its ability to create a consistent UI across different devices. The interface includes user authentication, a profile setup for health data, and a dedicated advice page.
  • Machine Learning Pipeline: The core prediction engine is built using Python’s scikit-learn and pandas.
  • Feature Engineering: The model considers variables such as age, gender, location (specifically population density/urbanization), smoking status, asthma history, and the current month.
  • Algorithm: The authors utilized a pipeline consisting of PolynomialFeatures (degree 2) followed by a Ridge Regression model. This approach allows the system to capture non-linear interactions between features—for instance, how the combination of "smoking" and "high pollen" might exponentially increase risk compared to either factor alone.

Datasets

The model was trained and evaluated using a set of fictional allergy data. While the paper does not utilize a massive public health database, it uses this synthesized dataset to establish a proof-of-concept for the multi-factor prediction logic. The data includes categorical mappings (e.g., 0 for no asthma, 1 for asthma) and numerical scales for environmental factors.

Key Insights and Ideas

  • The Contextual Gap: One of the most significant insights is the "interpretation bias." The authors found that users often perceive risk numbers differently (e.g., is a '3' low or moderate?). This led to the conclusion that a raw AI score is insufficient without accompanying qualitative guidance.
  • The "Text vs. Number" Hurdle: A recurring theme in the paper is the difficulty of mapping human-readable input (like a city name) to AI-understandable features (like a population density score). The authors highlight that for an AI to be effective in a mobile context, there must be a robust "translation" layer that handles user input validation.
  • Personal Motivation as a Driver: The research was heavily influenced by the authors' family experiences with severe allergies, emphasizing the need for "easy-to-use" tech that provides a "heads-up" before a user even leaves their house.

Key Results

The authors conducted two primary experiments to test the system:

  1. Accuracy Assessment: In a manual comparison between AI-generated scores and "true" risk assessments, the model showed promise but also revealed "blind spots." For example, in some test cases, the AI predicted a risk of 4/10 when the actual risk was closer to 8/10, suggesting that the current feature set needs to be more comprehensive.
  2. Input Validation Testing: The second experiment focused on how the AI handles text. It confirmed that the current iteration of the model is strictly numeric; entering "New York" instead of a population density code caused the system to fail. This highlighted the critical need for better UI-to-Model data mapping.

Limitations and Future Work

The authors are transparent about the app's current constraints:

  • Data Input Friction: Currently, users must enter numbers (like "3" for a specific location type) rather than natural language. Future work involves integrating APIs that can automatically pull population and pollen data based on a simple city name.
  • Model Refinement: The reliance on fictional data is a limitation. The authors plan to incorporate real-world datasets to improve the accuracy of the Ridge Regression model.
  • UX Improvements: The current version lacks a "front page" or onboarding flow. Future iterations will focus on a more intuitive user journey, moving the risk assessment to a more prominent position immediately after data entry.
  • Feature Expansion: There is a plan to include more granular environmental data, such as specific types of catkins or dust levels, to make the "Allergy App" a more robust tool for various respiratory sensitivities.

Summary updated

Abstract

This research paper introduces the mobile application called "Allergy App," designed to predict and manage pollen allergies using Artificial Intelligence (AI) and Data Science [1]. Allergies are widespread, often leading to serious health issues, such as asthma. The Allergy App tries to provide users with personalized risk assessments based on environmental and personal data, including location, age, smoking habits, and asthma history. The app's three key components include AI risk prediction, user authentication, and medical advice. There were several challenges during the development process and with the server [2]. The paper also shows experiments to assess the AI model's accuracy and address user input validation issues. While acknowledging some limitations, especially with text input formats, the Allergy App is a good tool for people to manage their allergies and reduce the risk of reactions, ultimately improving their quality of life.
BibTeX
@misc{liu2023intelligent,
  author = {Chengyu Liu and Julian Avellaneda},
  title = {{An Intelligent Detection and Analyzing Mobile Platform for Pollen Allergies Prediction using Artificial Intelligence and Data Science}},
  year = {2023},
  doi = {10.5121/csit.2023.131905},
}
Figure