[WEEK 5-Country Classification Using House Photos]
Team Members: Meltem Tokgöz , Enes Furkan Çiğdem , Asma Aiouez
This week in our project, we evaluated the results of the data in the single layer neural network code.For some hyperparameters we have tried, we have drawn confusion matrices and loss graph.Some of these are shown below.
We analyzed the results obtained from from these charts and matrices, our data is insufficient hence the accuracy is low.By considering the confusion matrix of images with 500 epoch and a learning rate of value 0.1 with 32x24 sizes, similarity is obtained between German and Turkey house images. Because of this, there is so much mispredictions.1000 epoch 0.01 learning rate with 48x48 size Confusion matrix is like previous situation , in this matrix,similarity between Indonesian and china houses .Thus we got low accuracy.The results are as shown in the table below.
We decided to increase the number of dataset this week to improve our results.We have increased the number of countries to 15.Also increased the number of image dataset each country.
We believe that this study will have a positive effect on our accuracy.
Secondly, we decided to use an different algorithm to give better results.
Now we’re going to try our datasets with cnn instead of the single layer neural network.
We also wrote a progress report for our project this week, and we sent it. In short up to the present, we’ve taken our first results and analyzed them. We’ll try the aforementioned paths to improve the results we taken next week.
