Multi-Layer Perceptrons to Classify HR, IRIS, & MNIST Data

Tech Stack: Python (scikit-learn - MLPClassifier & preprocessing, TensorFlow/Keras - 'adam' optimizer, numpy, matplotlib, seaborn)
Github URL: Project Link

This project explores the power and versatility of Multi-Layer Perceptrons (MLPs)—a class of artificial neural networks—across three real-world classification problems: employee attrition (HR data), handwritten digit recognition (MNIST), and flower species identification (IRIS). MLPs are able to model complex, non-linear relationships, making them highly effective for predictive analytics and data-driven decision-making.

👥

HR Attrition Prediction

Goal: Predict whether an employee will leave based on features like salary, commute time, evaluations, and satisfaction.
Approach: Categorical data preprocessing, visualization, and MLPClassifier with two hidden layers.
Result: 95.17% accuracy

⑤

MNIST Digit Recognition

Goal: Classify handwritten digits (0–9) from image data.
Approach: Data normalization, sequential MLP with TensorFlow/Keras (two hidden layers, sigmoid activation), and model tuning.
Result: 94.05% accuracy

🌸

IRIS Species Classification

Goal: Identify iris flower species from four feature measurements.
Approach: Data visualization, MLPClassifier with two small hidden layers, 'lbfgs' solver, regularization.
Result: 97.37% accuracy

Key Insights: MLPs are powerful, flexible, and robust tools for a wide range of classification tasks, especially where data is not linearly separable. Effective preprocessing, model architecture, and tuning are crucial for high performance, as demonstrated by strong results on all three benchmark datasets.

Author: Manraj Singh
Contact: manraj23singh@gmail.com