ManisAI v2.0

ManisAI v2.0

A Mobilenetv3 AI model that had been developed to classify eight(8) types of kuih:

1. Kek Lapis
2. Kuih Lapis
3. Kuih Ubi Kayu
4. Onde-onde
5. Kuih Seri Muka
6. Kuih Talam
7. Kuih Ketayap
8. 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

Full Changelog

ManisAI v1.1

ManisAI v1.1

A Mobilenetv3 AI model that had been developed to classify six types of kuih:

1. Kek Lapis
2. Kuih Lapis
3. Kuih Ubi Kayu
4. Onde-onde
5. Kuih Seri Muka
6. Kuih Talam

This model was trained using a dataset of over 3500 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 2 hours 41 minutes.

Report of the model:

✅ Overall Accuracy: 0.2832

📊 Per-class metrics:

Class	Precision	Recall	F1
Kek Lapis	0.1053	0.0208	0.0348
Kuih Lapis	0.0612	0.0588	0.0600
Kuih Seri Muka	0.0000	0.0000	0.0000
Kuih Talam	0.2569	0.7327	0.3805
Kuih Ubi Kayu	0.3168	0.3232	0.3200
Onde-onde	0.1333	0.0200	0.0348

🎯 ROC AUC per class:

Class	AUC
Kek Lapis	0.5613
Kuih Lapis	0.5283
Kuih Seri Muka	0.3986
Kuih Talam	0.6559
Kuih Ubi Kayu	0.6183
Onde-onde	0.3686

📦 Macro Metrics:

Marco Precision	0.1456
Macro Recall	0.1926
Macro F1	0.1383
Macro AUC	0.4125

Full Changelog

ManisAI v1.0

ManisAI v1.0

A Mobilenetv3 AI model that had been developed to classify six types of kuih:

1. Kek Lapis
2. Kuih Lapis
3. Kuih Ubi Kayu
4. Onde-onde
5. Kuih Seri Muka
6. Kuih Talam

This model was trained using a dataset of over 3500 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 2 hours 41 minutes.

Report of the newest model:

✅ Overall Accuracy: 0.2808

📊 Per-class metrics:

Class	Precision	Recall	F1
Kek Lapis	0.1987	0.3125	0.2429
Kuih Lapis	0.0000	0.0659	0.1127
Kuih Seri Muka	0.0000	0.0000	0.0000
Kuih Talam	0.0000	0.0000	0.0000
Kuih Ubi Kayu	0.1708	0.7677	0.2794
Onde-onde	0.9592	0.2883	0.4434

📦 Macro Metrics:

Marco Precision	0.2860
Macro Recall	0.2391
Macro F1	0.1797

🎯 ROC AUC per class:

Class	AUC
Kek Lapis	0.6803
Kuih Lapis	0.6815
Kuih Seri Muka	0.5940
Kuih Talam	0.7225
Kuih Ubi Kayu	0.6219
Onde-onde	0.4397