🚀 isrmltoolkit — Outlier Detection & Treatment for ML

isrmltoolkit is a Python toolkit designed to detect, handle, and visualize outliers in real-world datasets. Built for data scientists and ML practitioners, it combines statistical methods and machine learning algorithms into a simple, unified workflow.

• ✅ Clean your data
• ✅ Improve model performance
• ✅ Understand anomalies

---

🧠 Core Methods

📊 1. Interquartile Range (IQR)

A robust statistical method based on data dispersion, resistant to extreme values.

$$IQR = Q_3 - Q_1$$ $$\text{Lower Bound} = Q_1 - 1.5 \cdot IQR$$ $$\text{Upper Bound} = Q_3 + 1.5 \cdot IQR$$

👉 Values outside these bounds are flagged as outliers.

📐 2. Z-Score

Measures how far a value deviates from the mean in terms of standard deviations.

$$Z = \frac{X - \mu}{\sigma}$$

• X: Observed value
• $\mu$: Mean
• $\sigma$: Standard deviation

👉 Typically, $|Z| > 3$ indicates an outlier.

🧮 3. Mahalanobis Distance

Captures multivariate outliers by considering feature correlations.

$$D^2 = (X - \mu)^T \Sigma^{-1} (X - \mu)$$

• $\Sigma^{-1}$: Inverse covariance matrix

👉 High distance values indicate anomalous observations in multi-dimensional space.

🌲 4. Isolation Forest

A tree-based ML algorithm that isolates anomalies instead of profiling normal data.

• Uses random feature splits.
• Fewer splits $\rightarrow$ higher anomaly likelihood.
• 👉 Highly efficient for high-dimensional datasets.

🔧 5. Winsorization

Reduces the impact of extreme values without removing them by capping them at specific percentiles.

$$X' = \begin{cases} P_{low} & \text{if } X < P_{low} \ X & \text{if } P_{low} \le X \le P_{high} \ P_{high} & \text{if } X > P_{high} \end{cases}$$

👉 Useful for stabilizing distributions and improving model robustness.

---

⚡ Why isrmltoolkit?

• 🧠 Multiple Strategies: Diverse outlier detection methods in one place.
• ⚡ Real-world Ready: Designed for noisy, complex datasets.
• 📊 Pipeline Friendly: Built for both exploratory analysis and preprocessing.
• 🚀 Evolving: Actively updated with new features and algorithms.

---

🚀 Installation

pip install isrmltoolkit