This repository was archived by the owner on Nov 22, 2025. It is now read-only.
ManisAI v2.0
A Mobilenetv3 AI model that had been developed to classify eight(8) types of kuih:
- Kek Lapis
- Kuih Lapis
- Kuih Ubi Kayu
- Onde-onde
- Kuih Seri Muka
- Kuih Talam
- Kuih Ketayap
- Kuih Kaswi Pandan
This model was trained using a dataset of over 5000 images sourced from the internet, user-submitted photos, and a kuih dataset from Kaggle, and it is trained under Google Colab Notebok's T4 GPU for 6 hours 39 minutes.
Report of the model:
✅ Overall Accuracy: 0.9974
📊 Per-class metrics:
| Class | Precision | Recall | F1 |
|---|---|---|---|
| Kek Lapis | 0.9897 | 1.0000 | 0.9948 |
| Kuih Kaswi Pandan | 0.5000 | 1.0000 | 0.6667 |
| Kuih Ketayap | 1.0000 | 1.0000 | 1.0000 |
| Kuih Lapis | 1.0000 | 1.0000 | 1.0000 |
| Kuih Seri Muka | 1.0000 | 1.0000 | 1.0000 |
| Kuih Talam | 1.0000 | 1.0000 | 1.0000 |
| Kuih Ubi Kayu | 1.0000 | 0.9899 | 0.9949 |
| Onde-onde | 1.0000 | 0.9969 | 0.9985 |
🎯 ROC AUC per class:
| Class | AUC |
|---|---|
| Kek Lapis | 1.0000 |
| Kuih Lapis | 1.0000 |
| Kuih Seri Muka | 1.0000 |
| Kuih Talam | 1.0000 |
| Kuih Ubi Kayu | 1.0000 |
| Onde-onde | 1.0000 |
📦 Macro Metrics:
| Marco Precision | 0.9362 |
| Macro Recall | 0.9984 |
| Macro F1 | 0.9569 |
| Macro AUC | 1.0000 |