Deep Learning · Keras

ANN Seeds Classification

Neural-network classifier built for the Seeds dataset to distinguish three wheat varieties. The project evaluates architecture choices and regularization behavior, then explains model outcomes with learning curves and confusion-matrix diagnostics.

Dataset

Tabular data with 7 normalized morphological features and 3 classes.

Core Stack

Python, Keras/TensorFlow, NumPy, Scikit-learn, Matplotlib.

Architecture

Dense 64-32 + Softmax output, Adam optimizer, multiclass setup.

Experiment

Performance comparison between no-dropout and dropout (0.3).

Model Design

Input layer with 7 standardized numerical attributes.
Two hidden layers (64 and 32 neurons) using ReLU activation.
Softmax output for 3-way classification.
Training setup: 80 epochs, batch size 16, Adam (learning rate = 0.001).

Result summary: the no-dropout configuration converged faster and delivered the strongest overall performance for this dataset.

Evaluation Visuals

Accuracy comparison between ANN models — Accuracy comparison for dropout vs no-dropout configurations.

Loss comparison between ANN models — Training and validation loss behavior across epochs.

Confusion matrix for wheat seed classifier — Confusion matrix confirms strong class-level discrimination.

Request Project Files Back to Portfolio

Project package is shared on request to keep the portfolio lightweight.