Template Credit: Adapted from a template made available by Dr. Jason Brownlee of Machine Learning Mastery.
SUMMARY: This project aims to construct a predictive model using a TensorFlow convolutional neural network (CNN) and document the end-to-end steps using a template. The Intel Image Classification dataset is a multi-class classification situation where we attempt to predict one of several (more than two) possible outcomes.
INTRODUCTION: This dataset contains over 17,000 images of size 150×150 distributed under six categories: buildings, forest, glacier, mountain, sea, and street. There are approximately 14,000 images in the training set and 3,000 in the test/validation set. This dataset was initially published on https://datahack.analyticsvidhya.com by Intel as part of a data science competition.
From iteration Take1, we constructed a simple three-layer CNN neural network as the baseline model. We plan to use this model’s performance as the baseline measurement for future iterations of modeling.
In this Take2 iteration, we will construct a VGG16 neural network as an alternate model. We will compare this model’s performance with the baseline model from iteration Take1.
ANALYSIS: From iteration Take1, the baseline model’s performance achieved an accuracy score of 88.62% after 30 epochs using the training images. The baseline model also processed the validation images with an accuracy score of 85.37%.
In this Take2 iteration, the VGG16 model’s performance achieved an accuracy score of 83.57% after 30 epochs using the training images. The VGG16 model also processed the validation images with an accuracy score of 79.53%.
CONCLUSION: In this iteration, the TensorFlow CNN model appeared to be suitable for modeling this dataset. We should consider experimenting with TensorFlow for further modeling.
Dataset Used: Intel Image Classification Dataset
Dataset ML Model: Multi-class image classification with numerical attributes
Dataset Reference: https://www.kaggle.com/puneet6060/intel-image-classification
One potential source of performance benchmarks: https://www.kaggle.com/puneet6060/intel-image-classification
The HTML formatted report can be found here on GitHub.