Silvermind

Image from internet predicted
Any random 15 validation images predicted

Inspiration

As a country with ageing population, and therefore an immense need to care for the seniors, the moment we deduced that we wanted to work with drugs, because of its potential, we decided that our problem statement was the path to take, as we felt that this was an issue faced by seniors. The problem that we tackle is seniors mixing up the tablets and taking the wrong medication at wrong times/dosages as they misplace the packaging or manual, which is quite prevalent. It is in fact an issue even amongst youth and adults these days. Using this code, we would like to tackle this issue.

What it does

It predicts and classifies images of pills. From the predicted pill image, it will look through the database of pills to find relevant information about the pill to answer queries from the user such as "What is the usage of this medicine?". Then, it will provide lifestyle advice according to the medicine they consume to aid with their help using OpenAI.

How we built it

We used Keras from Tensorflow to build a Convolutional Neural Network (CNN). It begins with convolutional layers followed by batch normalization and max pooling layers, which help extract meaningful features from the input data. Dropout layers are introduced to prevent overfitting. The flattened output is then passed through dense layers for classification. The model has a total of 12 layers, including convolutional, batch normalization, max pooling, dropout, and dense layers.

Challenges we ran into

Our model initially was overfitted and could not generalize the data well. Hence, the predictions were not accurate. Hence, we added dropout in our model to prevent overfitting and fine-tuned the model to make it more accurate. It was also challenging to implement the prediction of the images from the internet as we had to do some preprocessing.

Comments on the model

Our model has an accuracy of 0.9195 and the difference in training model and validation model accuracy is 0.0008 which is a good indicator that it can generalize well.

What we learned

We learnt how to apply the knowledge taught in the workshops to create a meaningful model that can make a difference to the lives of people around the world
We learn how to coordinate and work together well as a team.