> For the complete documentation index, see [llms.txt](https://saral.sunbird.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://saral.sunbird.org/learn/ml-model-accuracy-results.md).

# ML Model Accuracy/Results

## Handwritten Alphanumeric model

#### Iteration 1:

| Model details                  | Dataset          | Accuracy achieved |
| ------------------------------ | ---------------- | ----------------- |
| Model trained and inference on | Existing dataset | 99.80%            |
| Model inference on             | NIST dataset     | 62.70%            |
| Average across all datasets    | -                | 81.25%            |

**Reasoning** - As model is trained on the existing dataset, doesn't perform good on NIST dataset

#### Iteration 2:

| Model details    | Dataset                                    | Accuracy achieved |
| ---------------- | ------------------------------------------ | ----------------- |
| Model trained on | Existing dataset + NIST misclassifications | 83.30%            |

**Reasoning** - After training the model on NIST misclassifications, improvement in accuracy

#### Iteration 3:

| Model details    | Dataset                                    | Accuracy achieved |
| ---------------- | ------------------------------------------ | ----------------- |
| Model trained on | Existing dataset + NIST misclassifications | 93.90%            |

**Reasoning** - After training the model on NIST misclassifications as per previous checkpoint, improvement in accuracy

#### Iteration 4:

| Model details    | Dataset                                                                  | Accuracy achieved |
| ---------------- | ------------------------------------------------------------------------ | ----------------- |
| Model trained on | Existing dataset + manually collected dataset from sheets (\~1K samples) | 93.90%            |

**Reasoning** - Training the model upon previous checkpoint and adding manually collected data, improvement in accuracy

## Handwritten Digits model

#### Iteration 1:

| Model details                  | Dataset                                 | Accuracy achieved |
| ------------------------------ | --------------------------------------- | ----------------- |
| Model trained and inference on | Existing dataset                        | 99.90%            |
| Model inference on             | NIST dataset                            | 60.00%            |
| Model inference on             | Obtained production data (\~50 samples) | 97.70%            |
| Average across all datasets    | -                                       | 85.80%            |

**Reasoning** - As model is trained on the existing dataset, doesn't perform good on NIST dataset

#### Iteration 2:

| Model details    | Dataset                                   | Accuracy achieved |
| ---------------- | ----------------------------------------- | ----------------- |
| Model trained on | Existing dataset + NIST misclasifications | 96.40%            |

**Reasoning** - After training the model on NIST misclassifications, improvement in accuracy

#### Iteration 3:

| Model details    | Dataset                                                                       | Accuracy achieved |
| ---------------- | ----------------------------------------------------------------------------- | ----------------- |
| Model trained on | Existing dataset + NIST misclasifications + production dataset (\~50 samples) | 99.70%            |

**Reasoning**: After training the model on NIST misclassifications and production dataset, improvement in accuracy

#### Iteration 4:

| Model details    | Dataset                                                                                             | Accuracy achieved |
| ---------------- | --------------------------------------------------------------------------------------------------- | ----------------- |
| Model trained on | Existing dataset + NIST misclasifications + manually collected dataset from sheets (\~8.6k samples) | 98.30%            |

**Reasoning**: As averaging upon a large production dataset, the accuracy slightly dips as compared to iteration 3

## Sample dataset images

#### Existing dataset

**Handwritten alphanumeric**

![0a56f9eb1545428f8d9aea0665b7c460](https://user-images.githubusercontent.com/104554231/229728834-a7681ad6-bd4f-488b-8e65-62052d418a02.jpg) ![0b0de241b7e64e918fada2f896efdf24](https://user-images.githubusercontent.com/104554231/229728866-6c7b8765-3cb9-4acb-b99e-6c3b1e3cf4bf.jpg) ![00b6ef24818b4323b4ada81ad433af67](https://user-images.githubusercontent.com/104554231/229728911-c39fd804-f2ec-4dfa-9075-9d64e7c84fa1.jpg) ![0ac132e3693c47ccb2fe40fb465ba060](https://user-images.githubusercontent.com/104554231/229728951-b2815ce4-f177-40ca-8192-c6eb2778c04e.jpg) ![0aeaa24048ff4c5d8312d37671c6e08e](https://user-images.githubusercontent.com/104554231/229728980-9a5a1fbb-7644-49c8-b2fa-57a974a69e0e.jpg) ![0b7f1f188faf453b83704a069791f28f](https://user-images.githubusercontent.com/104554231/229729018-546e5cb8-7cd1-40ee-b92a-24660bd7eb50.jpg) ![0bd15fe615db458781bce7dc32a926b5](https://user-images.githubusercontent.com/104554231/229729053-75efe9dd-7769-4595-8dce-faac1839bd8d.jpg) ![0adb585eae814b62a56695650cac0666](https://user-images.githubusercontent.com/104554231/229729111-a142f1a1-22bc-43ac-9a04-11f77fabe794.jpg) ![0b0ffa9c3dcf497c8da259d391f5f159](https://user-images.githubusercontent.com/104554231/229729165-b4060026-56b8-4b1c-ace2-f6d1d17e403e.jpg) ![0b2c4c540a7247629e1dd7a9abd59962](https://user-images.githubusercontent.com/104554231/229729203-2e311e0d-e77a-4076-a338-be7c2ad58e48.jpg)

**Handwritten digits**

![0b1a2b29b0904989a3f6df3a13b93d89](https://user-images.githubusercontent.com/104554231/229729405-0ac01786-e0b8-4724-90c9-f94b7da24581.jpg) ![0abb9b222a1142bdaabd30b6742ca8b2](https://user-images.githubusercontent.com/104554231/229729424-baee18ff-46b5-4612-8fd2-0f12bda1d658.jpg) ![0afd4ac792f14b8185d54cb9d0937b3e](https://user-images.githubusercontent.com/104554231/229729467-fea6c80e-f714-4e8b-882e-50f554338779.jpg) ![0ac2ad40-7af8-44b4-9829-284a2b33416c\_printed](https://user-images.githubusercontent.com/104554231/229729496-2c3eb05b-3f22-44b6-9c81-7fc0d7c9d4c9.jpg) ![00af4fd32ddc42efbfb1222725599ecf](https://user-images.githubusercontent.com/104554231/229729524-9d6f9b28-6689-452f-a6fe-53d2a28c04ee.jpg) ![0aa35cc5-4472-455d-a575-7167c15df849\_generated](https://user-images.githubusercontent.com/104554231/229729574-0ce01249-f857-49b5-8c68-bebe392fc562.jpg) ![0a8986b29e804002a3136ddf110f3638](https://user-images.githubusercontent.com/104554231/229729638-552ef113-a3c6-447f-abe7-c6c5fdd46ce2.jpg) ![0a1d909f2f2847b795ac40cefa452c3e](https://user-images.githubusercontent.com/104554231/229729688-4fcec72c-fc81-4be6-9800-b725b69e4f1f.jpg) ![0\_\_0040c46a-eae7-4219-bffb-7dd418ad9ffb\_up\_govt](https://user-images.githubusercontent.com/104554231/229729726-92b53d2b-bb33-4e1a-8324-25b880ad56ff.jpg) ![00c9de07f7b640d083680a5c21661a8b](https://user-images.githubusercontent.com/104554231/229729771-5b7fd3fd-404b-45c5-bcc4-6e16b21ad20d.jpg)

#### NIST dataset

**Handwritten alphanumeric**

![hsf\_0\_00017](https://user-images.githubusercontent.com/104554231/229730299-9d3eb74d-6824-4f05-bfbc-a7ef6a2c9257.png) ![hsf\_0\_00043](https://user-images.githubusercontent.com/104554231/229730351-eef9aae8-492f-4847-8a84-23bc733576e0.png) ![hsf\_0\_00017](https://user-images.githubusercontent.com/104554231/229730382-02997f2e-e4c1-4c0f-9e4d-e2793b2a4d90.png) ![hsf\_0\_00020](https://user-images.githubusercontent.com/104554231/229730415-a14cc5e9-5f27-48ae-b6e4-de00adc4c0af.png) ![hsf\_0\_00010](https://user-images.githubusercontent.com/104554231/229730469-ad835be3-79dc-4711-9b24-fed9f2b6eb43.png) ![hsf\_0\_00017](https://user-images.githubusercontent.com/104554231/229730508-87491dc2-d789-499d-bc50-4cfe564101c6.png) ![hsf\_0\_00020](https://user-images.githubusercontent.com/104554231/229730556-4d3d3834-4c23-4359-8e0f-2832a5cf89be.png) ![hsf\_0\_00008](https://user-images.githubusercontent.com/104554231/229730590-e064eb19-020c-4285-ada4-b074f22bd787.png) ![hsf\_0\_00025](https://user-images.githubusercontent.com/104554231/229730636-2e7131fd-b6c4-4784-a01c-fdbc3afd49d6.png) ![hsf\_0\_00010](https://user-images.githubusercontent.com/104554231/229730689-a3def9d9-0462-4c80-ad81-1ac4659f2ed4.png)

**Handwritten digits**

![hsf\_0\_00015](https://user-images.githubusercontent.com/104554231/229730804-f4cf3b8a-1676-4c59-aafe-40ad101701c2.png) ![hsf\_0\_00010](https://user-images.githubusercontent.com/104554231/229730833-fdd7070e-b2ad-4a93-9aec-8be4dd8ed0d1.png) ![hsf\_0\_00017](https://user-images.githubusercontent.com/104554231/229730849-48c5ae43-e46a-4a77-8a16-898b1b74d21f.png) ![hsf\_0\_00009](https://user-images.githubusercontent.com/104554231/229730879-b3eeab31-185e-488d-ad11-5d7f36e0b64f.png) ![hsf\_0\_00117](https://user-images.githubusercontent.com/104554231/229730925-14c9b2b2-385f-4f4d-900c-22ea831ca552.png) ![hsf\_0\_00009](https://user-images.githubusercontent.com/104554231/229730959-b5e501af-eda4-43b4-84d9-5aad992ba148.png) ![hsf\_0\_00012](https://user-images.githubusercontent.com/104554231/229730998-b65b66fb-99a6-450d-ac69-a983bc3f4395.png) ![hsf\_0\_00019](https://user-images.githubusercontent.com/104554231/229731041-a4a7dae7-3829-402b-960a-ffc75569dcdf.png) ![hsf\_0\_00009](https://user-images.githubusercontent.com/104554231/229731085-130d9b21-bae1-4e90-b328-26e70a8c498a.png) ![hsf\_0\_00013](https://user-images.githubusercontent.com/104554231/229731151-6a285899-bad4-43a2-87a7-863809a69afb.png)

#### Manually collected

**Handwritten alphanumeric**

![img014-032](https://user-images.githubusercontent.com/104554231/229731461-842be1d2-7f05-4b46-a9a9-7c33773ae1e6.jpg) ![img015-044](https://user-images.githubusercontent.com/104554231/229731590-65a8bf93-cce0-404d-a106-75ce0e9be53e.jpg) ![img016-045](https://user-images.githubusercontent.com/104554231/229731652-9e0980c2-7560-4267-a5cf-3fb09fae0b73.jpg) ![img023-039](https://user-images.githubusercontent.com/104554231/229731690-82841cf8-9bc6-4bd8-8ee8-8f0efbbc7d09.jpg) ![img026-001](https://user-images.githubusercontent.com/104554231/229731742-997c7dfd-bdd4-4776-bffb-5847c189b68e.jpg) ![img029-045](https://user-images.githubusercontent.com/104554231/229731791-d463c5fb-1f2f-490d-9f37-248146334d52.jpg) ![img013-040](https://user-images.githubusercontent.com/104554231/229731828-c6bc84e5-6f45-4182-aa67-b7abb6972212.jpg) ![img032-053](https://user-images.githubusercontent.com/104554231/229731876-b87e2d75-6e3e-4900-9701-9d8a071b0904.jpg) ![img034-055](https://user-images.githubusercontent.com/104554231/229731969-69e1e111-b13d-4f51-83b3-ee89dd384f02.jpg) ![img011-033](https://user-images.githubusercontent.com/104554231/229732030-4c5265ed-c320-481d-925b-1bd42eda74d8.jpg)

**Handwritten digits**

![33665](https://user-images.githubusercontent.com/104554231/229732166-dfb1e6f4-79a4-46dd-aa86-5c1eba6c6ba8.png) ![33679](https://user-images.githubusercontent.com/104554231/229732203-f38acb46-7212-4cc2-9718-15af59ff4c84.png) ![33758](https://user-images.githubusercontent.com/104554231/229732228-97da7a10-36b3-4c27-bdb4-38ec83bd112a.png) ![33745](https://user-images.githubusercontent.com/104554231/229732258-a04a6986-902c-471f-920a-0ca86639cca6.png) ![33650](https://user-images.githubusercontent.com/104554231/229732280-f7ef258a-3445-43f3-aa20-7ae8b57c1919.png) ![33753](https://user-images.githubusercontent.com/104554231/229732313-6e6841df-6024-4d68-affa-b50fea353fea.png) ![31372](https://user-images.githubusercontent.com/104554231/229732377-385301ff-6d24-4f76-b111-45b37caf7596.png) ![33362](https://user-images.githubusercontent.com/104554231/229732444-23a8ef8a-0dd1-4789-846f-d675f2b75e9b.png) ![31075](https://user-images.githubusercontent.com/104554231/229732556-37b99ce5-4e90-4e62-8c69-c6162cdc789d.png) ![31065](https://user-images.githubusercontent.com/104554231/229732609-20634d4f-a4eb-43e2-b03f-a7ac0425dc02.png)

#### Some unhandled misclassifications

![18103](https://user-images.githubusercontent.com/104554231/229733361-616dc17a-9c84-4d3f-9046-8790fc4009b6.png) ![28926](https://user-images.githubusercontent.com/104554231/229733480-2d30388b-1653-4778-bda0-4949899ea6f2.png) ![607](https://user-images.githubusercontent.com/104554231/229733515-47e25ccf-8448-49bd-8456-b65acf300dd6.png) ![30658](https://user-images.githubusercontent.com/104554231/229733543-1604df90-5953-4b3b-9312-f3759e924ca6.png) ![402](https://user-images.githubusercontent.com/104554231/229733591-6b75ac87-e14c-4c13-9035-3fcd8f510e69.png) ![32358](https://user-images.githubusercontent.com/104554231/229733649-ac697ba2-7c1a-4e56-bceb-dff83a5f684e.png) ![24141](https://user-images.githubusercontent.com/104554231/229733812-ec789498-8f13-4374-8f1d-c9aaf1f8a0ed.png) ![30838](https://user-images.githubusercontent.com/104554231/229733864-cee75918-1478-4a2f-8443-dd7fe3585430.png) ![24158](https://user-images.githubusercontent.com/104554231/229733937-99a629bd-a961-497d-af0b-8e5c87f4f97f.png) ![32225](https://user-images.githubusercontent.com/104554231/229733986-c1bf94b8-8efe-438f-a737-6d4c99725d48.png)

**Reasoning**: Generally occurs if the digits are written in corners of the cell


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://saral.sunbird.org/learn/ml-model-accuracy-results.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.