Verification of an Unbiased Inverse Square-Root Transformation with Linear Regression

Validated a proposed transformation technique using stock data to enhance model linearity and prediction accuracy.

---

Overview

Evaluated whether a custom inverse square-root transformation improves regression performance on financial datasets (N225).

---

Steps

1. Data Collection
   • Merged 9 different data sources and web-scraped 25,000+ stock entries.
2. Preprocessing
   • Cleaned missing values, normalized distributions, verified normality assumptions.
3. Modeling
   • Compared classical vs unbiased inverse-square-root transformed features.
4. Evaluation
   • Measured R², MSE, MAE improvements.