/AI-Creator

Primary LanguagePythonMIT LicenseMIT

AI Creator

license issues last commit downloads

Table of contents

Preface

With this program, we tried to predict heart failure. We did this by creating a program that can train an AI and then use it to predict unknown data. In the Netherlands, it is common to make a PWS in the last year of high school. The PWS is a big assignment where you get to choose how you want to fill it in. We wanted to combine biology and informatics, but we didn't know-how. After some thinking, we finally figured out how to combine biology and informatics. We came up with the following idea: to make an AI that can predict whether heart failure is likely. To accomplish this, we divided this project into two. One person will do the programming, the other one will be the data analyst. We think we've made something with huge potential, something that will be very useful in the future.

We hope you enjoy checking out our work!

Introduction

Heart failure is one of the most common causes of death. Those affected often aren't aware they are at risk. With this program, we try to show people whether it is likely that they will develop heart failure. Hopefully, the program can stimulate more people to go see a doctor when necessary. This might result in fewer deaths caused by heart failure.

Description of what is already known

What is an AI

Artificial Intelligence (AI) is nothing more than a computer trying to imitate human intelligence. AI includes Machine Learning and deep learning. Both can be used for different purposes. For example, deep learning will work better if more data is available. Processing a large amount of data requires a high-end machine.

A machine learning model consists of neurons and weights. While learning, the computer will adjust these weights to get higher accuracy. The neurons are divided into three categories the input layer, hidden layer, and the output layer. The input layer is your input values, and the output layer is the prediction. The hidden layers can be adjusted to get better accuracy.

For more information on how a neural network works click here

Own research

To know more about heart failure, we did some research on the heart. How to heart functions and how heart failure can arise. To read more about the research we did, click here

Experiments

results.png

We discovered that most values in the dataset have some impact on the result but the best input values are age, gender, ap_hi, ap_lo, cholesterol, and gluc. Also when we removed one of these values as a test and we discovered that this would drastically increase the MAE (mean absolute error).

Product

We made a program that can create and train a model and then predict unknown data

Installation

  • Installing using the exe installer isn't available with the current release. Check release details for more info.
  • Download the latest release (you can use the zip or exe file)
  • Extract the zip file on your computer or install the program using the exe file
  • Find the installation folder or where you extracted the zip file and open GUI.exe

How to use

  • Open the file named 'GUI.exe' -- If the program gives an error, find the settings.json file open it and save it.
  • When the program starts for the first time you need to select the dataset (csv file) you want to use. (This can be one of the datasets given below or your own dataset)
  • This opens another window where you can update the minimum and maximum values. If you don't want to do this select the label and click on add. If there are no errors close the window and the program starts up.
  • To create an AI go to File and select New File
  • Here you can set different parameters for the AI
  • When the parameters are set, you can train the AI by going under Train and selecting Train
  • When done, you can predict with unknown data and save the file located under the File section
  • If you want more information on how to use the program see the Wiki

Available datasets

Settings

  • In settings.json are all the available datasets listed and it is possible to add a dataset yourself.
  • You can do this by sorting all the lines. I use a VS Code extension for this, and then add your dataset with the input_values, label, and mapping
  • To change the default save folder and the dataset you can either edit the settings.json or inside the application under Settings and then Other Settings.

Conclusion

We made a functioning AI that can predict heart failure. However, the AI is very basic and only uses 10 different variables in its calculations. When we take a look at the results of the experiments we've done, it becomes clear that even though we change some input values the average stays roughly the same. When we change the hidden layers we see that larger hidden layers improve the average result but with this method, it takes longer to create a model. It also doesn't greatly improve the average result, only by a small margin. It is interesting to see that height and weight are not improving the results. But when you think about it, it isn't weird. Because in the dataset we used there is a value cholestorol, height and weight aren't important anymore. Cholestorol is also a known factor of causing heart failure. A note: the AI does NOT replace a doctor at all. If you think you may be at risk for heart failure, go see a doctor.

Reflection

The process of this project went very smoothly. Although we ran into a couple of issues. Firstly, bugs, we experienced lots of them. (more in detail explanation further down) Secondly, when the AI was transferring every single value into a value it can understand. This resulted in data nodes that contained numbers like 1,2,3,4 being interpreted by the program as 4 is better than 1 or 1 is better than 4. This was in some cases not good so to improve the accuracy of the AI, we made an option in the program to split those values into data nodes. For example, one data node contained four values (1,2,3,4) this was transferred to four data nodes each containing zero or one (true or false).

Bugs

Every programmer knows that bugs come with developing a program, but some bugs could have been prevented. Like the bugs where the naming of the variables was the reason. As an example, there was some confusion when changing some variables in the program. Also, there were some large bugs, some resulting in the datasets that couldn't be opened but this was easily resolved when the issue was found.

Credits

We would like to thank Mr. Postma for guiding us through the process of building this program. In addition, we want to thank Mr. Gijsbers for providing us with useful information about the heart. Finally, we want to thank Dinand Wesdorp for his help in the development of the GUI.

Citation