In my opinion, this is a question on model's reliability and availability. We say a system is available if it performs the assigning work. On the other hand, a system is said to be reliable if it performs the assigned work correctly. This principle is applicable to most real-world scenarios.

For example, assume that a person makes a withdrawal of Rs. 10000 from an ATM machine, and the machines dispatches amount worth Rs. 9000. The system cannot be said to be not performing, as it is producing a response by dispatching the money. However, the accuracy of the system is not satisfactory because the system did not produce correct response.

Specifically talking on machine learning models, in the current real-world scenarios, the systems/models with highest accuracies survive and are used globally. Models which simply perform and produce reasonable/medium accuracies are termed OTHERS.

Dear Kgs Venkatesan ,

I am a bioinformatician and using machine learning algorithms to analyze the data. According to my POV model accuracy is more important and its all depends on the training data.

Dear Professor Kgs Venkatesan :

In my humble opinion, accuracy is the most important characteristic of a ML/AI model. When we discuss the performance of a model, we need to first clarify whether it is the Model training performance or model scoring performance? Model performance can be improved using distributed computing and parallelizing over the scored assets, whereas accuracy has to be carefully built during the model training process.

Often times, when we compare the computational times for training a model with that of scoring the same model, the latter is several order of magnitudes lower. Furthermore, once models are deployed in operation, usually model (re)training is done much less frequently than scoring the outputs.

Dear,

this depends in the application and the field for example for real time application for obstacle detections we need both but for medical imaging application we need more the accuracy.