From Raw AI-Dependency Data to Robust Cognitive-Growth Prediction: A Machine Learning Framework

This repository contains the full analysis pipeline, code, and supporting materials for a study assessing the impact of artificial intelligence (AI) dependency on human cognitive growth using a machine learning framework applied to a simulated dataset of ~52,000 records.

Overview

Ten baseline classifiers (Logistic Regression, MLP, CatBoost, SVM, XGBoost, LightGBM, Random Forest, Extra Trees, Decision Tree, KNN) are compared for their ability to classify individuals into three cognitive-outcome categories — Cognitive Decline Risk, Stagnant/Moderate, and Positive Growth — based on demographic, behavioral, psychological, and AI-usage features.

Key components of the analysis:

• Exploratory data analysis (distributions, correlation structure)
• Feature engineering and multi-criteria feature importance ranking (RF importance, mutual information, chi-square, ANOVA F-score)
• Detection and reporting of a target-leakage concern in the Scenario feature
• Ten-classifier performance benchmarking (accuracy, precision, recall, F1, MCC, Cohen's kappa)
• Confusion matrix and ROC/AUC analysis
• Statistical comparison of classifiers (chi-square test of homogeneity; pairwise two-proportion z-tests with Holm–Bonferroni correction)

Repository Structure

.
├── README.md                  <- This file
├── LICENSE                    <- MIT license for code in this repository
├── CITATION.cff                <- Citation metadata (for "Cite this repository")
├── requirements.txt            <- Python dependencies
├── .gitignore
├── notebooks/
│   └── main_analysis.ipynb     <- Full pipeline (see Colab link below)
├── data/
│   └── README.md               <- Data access instructions (data not redistributed here)
├── results/
│   ├── figures/                <- Generated charts, confusion matrices, ROC curves
│   └── tables/                 <- Model comparison, feature ranking, statistical test tables
└── docs/
    └── paper/                  <- Manuscript-related material (optional)

Data Availability

The dataset used in this study is publicly available on Kaggle: https://www.kaggle.com/datasets/laveshjadon/ai-impact-on-human-mental-growth

The dataset is not redistributed in this repository. See data/README.md for download and setup instructions.

Code Availability

The full analysis pipeline is available as a Google Colaboratory notebook: https://colab.research.google.com/drive/1OeKT63E0cWmbD12y19lXH2v_Egms0XZi

A copy of the notebook is also maintained in notebooks/main_analysis.ipynb in this repository.

Getting Started

1. Clone the repository

git clone https://github.com/ahmedsal95/AI-Effect-on-Mental-Health.git
cd AI-Effect-on-Mental-Health

2. Install dependencies

pip install -r requirements.txt

3. Download the dataset

Download the dataset from Kaggle (link above) and place it in the data/ directory. See data/README.md for details.

4. Run the notebook

jupyter notebook notebooks/main_analysis.ipynb

Or open directly in Google Colab using the badge at the top of this README.

Key Findings (Summary)

• Logistic Regression achieved the highest accuracy (99.79%), followed by a statistically indistinguishable tier of MLP, CatBoost, SVM, XGBoost, and LightGBM (all >99.2%).
• The Scenario_Pessimistic feature showed anomalously high importance scores (orders of magnitude above other features), flagged as a likely target-leakage artifact of the simulation design rather than a genuine predictor.
• Individual-level behavioral and psychometric variables (e.g., Mental Growth Index, Innovation Rate, Problem-Solving Skills) were the dominant legitimate predictors; demographic/regional/professional variables contributed negligibly.
• Statistical comparison (chi-square test of homogeneity + Holm-corrected pairwise proportion tests) confirmed that classifiers cluster into statistically distinct performance tiers rather than following a strict linear ranking.

Citation

If you use this code or refer to this work, please cite:

Salaheldin, A. (2026). From Raw AI-Dependency Data to Robust Cognitive-Growth
Prediction: A Machine Learning Framework. [Manuscript in preparation].
Code: https://github.com/ahmedsal95/AI-Effect-on-Mental-Health

See CITATION.cff for structured citation metadata.

License

The code in this repository is released under the MIT License. See the data/README.md file for the dataset's own license terms (governed by Kaggle/the dataset creator, not by this repository).

Author

Ahmed Salaheldin Sole author — conceived and designed the study, performed data analysis, developed the machine learning models, interpreted the results, and wrote the manuscript.

Declarations

• Funding: This study did not receive any specific funding.
• Conflicts of Interest: The author declares no conflict of interest.
• Ethics Approval: Not required — this study used a publicly available, simulated/secondary dataset with no direct human-subject data collection.