Cyril Robinson Azariah J.

The reduction in accuracy when converting a machine learning model from Python to an Android app could be due to several reasons:

Several reasons you might see poor accuracy when testing an image using an Android app compared to testing the same image using a Python program. Here are a few possibilities:

It's difficult to determine the exact reason for the differences in accuracy without additional information, such as the specific image recognition algorithm used, the training data, and the implementation details.

There are several reasons why the accuracy of image classification in an Android app might be lower compared to using Python:

Data Preprocessing: In Android, you may not have access to the same image processing libraries that are available in Python. This can make preprocessing the image data more difficult and lead to lower accuracy.

Model Complexity: The available computational resources on an Android device are often limited compared to a desktop or laptop computer. This means that you may need to use a simpler model in an Android app, which can result in lower accuracy.

Model Optimization: The deep learning libraries used in Python may be optimized for performance on desktop and laptop computers, whereas the libraries used in an Android app may not be as optimized. This can result in slower performance and lower accuracy.

Model Size: The size of the model may also be a factor, as larger models may be too large to fit in the memory of an Android device. In such cases, you may need to reduce the size of the model, which can result in lower accuracy.

Janak Trivedi Md. Humayun Kabir Muhammad Harris Hashmi Thanks for your answers. I also learnt that these could be the reasons why the performance of an ML/DL model on an Android app is lower compared to when it's run in Python. Though, there could be several reasons, some of the common reasons are:

Muhammad Harris Hashmi as you mentioned it is due to "Model size". So, when deploying a model to an Android app, the size of the model is a critical factor. Larger models take longer to run on mobile devices, and their performance might also be affected.

The other main reason might be the "Computing power" which Janak Trivedi you have also mentioned. As we know, Mobile devices have limited computing power compared to laptops or cloud-based systems. This can lead to slower model performance and reduced accuracy.

Muhammad Harris Hashmi You also mentioned about the data Preprocessing, which I also agree. I understood, the data preprocessing steps performed on the images before feeding them to the model might be different between Python and the Android app. This can lead to discrepancies in the input data and result in reduced accuracy.

Janak Trivedi As you mentioned, the other reason is due to the "Overfitting". In the model, there might be overfitting the training data when trained on Python, but when deployed on the Android app, it might not generalize well to unseen data.

I thought "Data augmentation" improved the accuracy but In some cases, the model might be trained on a larger and more diverse dataset that includes data augmentation, whereas on the Android app, the model might not have access to the same data or preprocessing steps.

Md. Humayun Kabir You have very well listed the reasons which will impact the accuracy. After I surveyed some literature, I found that to improve the performance of the model, you might also wanted to consider some of the following steps:

• Reduce the size of the model: You can try to reduce the size of the model by removing some of the layers or using a more compact architecture.

• Data augmentation: Increasing the size of the dataset by adding augmented images and including different backgrounds can help the model generalize better.

• Fine-tune the model: You can try to fine-tune the model by training it on a smaller dataset that is more representative of the data that the model will encounter in the Android app.

• Preprocess the data: It's important to preprocess the data in the same way both in Python and on the Android app to ensure that the input data is consistent.

• Consider using transfer learning: Instead of training a model from scratch, you can consider using a pre-trained model and fine-tuning it for your specific task.

You also keep in mind that these steps might not guarantee an improvement in accuracy, and you might need to try multiple approaches to find the best solution for your problem. If I missed something, please mention in this comment.