[WEEK 5 — Predicting the pieces of furniture in living rooms]

Group Members: Mohammed ALI, Aybüke Yalçıner, HaticeAcar.

Last week, we worked on 3 different algorithms that are the convolutional neural network, multilayer perceptron and K — Nearest Neighbors. These algorithms are run on 10 different classes. These are Armchair, Carpet, Coffee Table, Lamp, Painting, Sofa, Coffee Table, Television, Television Unit and Vases.

We can see the accuracies of algorithms on all dataset in Figure 1. When we look at the table, we get the best accuracy with Convolutional Neural Network. So we take the 56.8 as our base accuracy. In addition, We cannot get any result from K — Nearest Neighbor for the multiclass classification because we have a lot of data so it takes too much time.

Figure 2: The structure of the Convolutional Neural Network

Figure 2 is our base convolutional neural network structure that we get 56.8 accuracies with it. We use 50 as batch size and 5 epochs.

This week, we try to evaluate our base accuracy which is 56.8 on test dataset that we got it with Convolutional Neural Network. We made some search to learn how we can evaluate this accuracy. And some steps we found as below:

1. Adding more layers
2. Some image processing skills
3. Changing convolution neural network parameters
4. Adding more hidden layers to the multilayer feed-forward neural network that is at the end of the neural network
5. Collecting more training data

These are some strategies, that we find, to get more accuracy. So we try to apply these ones by one and see the effects. On the other hand, we continue our research to gain more information.

References:

[1] https://towardsdatascience.com/image-classification-python-keras-tutorial-kaggle-challenge-45a6332a58b8

[2] https://www.quora.com/How-do-I-increase-accuracy-using-Convolutional-Neural-Networks-CNNs-ConvNets-for-regression

[3] https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html