For our project, I decided to build a machine learning model using Teachable Machine to create a special cat door that would only allow cats to pass through. The concept was simple: train the model to distinguish between cats and non-cats, such as other animals, so that only cats could gain access to the door. I gathered a dataset of images featuring various cats and non-cats, hoping to create a reliable model. I focused on capturing different types of cats and non-cats, trying to cover a range of angles, lighting conditions, and distances. However, the model didn’t perform as expected.
One of the main issues I encountered was that the model didn’t properly convert the images into usable output. Although Teachable Machine learned from the data I provided, it didn’t translate into a functional system that could effectively identify cats. I suspect this failure was due to insufficient or imbalanced data. If I had included more diverse images, including cats in different environments or under varying lighting conditions, the model likely would have learned more effectively. I also considered adding sound to the dataset, like meowing or purring, but didn’t have time to experiment with that.
Reflecting on what other data might have helped, I realize that having more images of cats from different breeds, along with more varied environments and backgrounds, would have contributed to the model’s ability to generalize better. If I had more time or resources, I could have incorporated video clips of cats to show them in motion. Sounds, such as a cat’s meow or even just the noise of a cat approaching the door, could have been useful as well. Having a more diverse range of data would likely have helped the model become more accurate.
Unfortunately, the model didn’t work as intended. There were several instances when it failed to recognize cats correctly, especially in situations where the images were low quality or when the cats were partially obscured. The model also struggled with differentiating between cats and other animals that resembled them, like small dogs or certain wildlife. These challenges highlighted the importance of having high-quality, diverse training data and the potential impact of environmental factors, like lighting or background noise, on the model’s performance.
Despite the model’s shortcomings, I believe that with more data, especially videos and sounds, it could become a more accurate tool for identifying cats. For now, I haven’t been able to generate a functional link to the project, but this experience has provided valuable insights into how to approach training a machine learning model and the types of data needed for a successful project.