https://github.com/harsh-git98/brain-dead-hackathon
https://github.com/harsh-git98/brain-dead-hackathon
Last synced: over 1 year ago
JSON representation
- Host: GitHub
- URL: https://github.com/harsh-git98/brain-dead-hackathon
- Owner: Harsh-git98
- Created: 2024-04-06T18:01:36.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2024-06-08T20:29:37.000Z (about 2 years ago)
- Last Synced: 2025-01-24T04:56:12.749Z (over 1 year ago)
- Language: Jupyter Notebook
- Size: 248 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Brain Dead Hackathon Submission Report
## Author:
- Harsh Ranjan
- Vanshika Kothari
- Sayan Roy
- Aaratrika Sarkar
- Harsh Raj Gupta
## Proposal: Analysis and Prediction of Rice Production using ARIMA and LSTM Models
### Problem Statement
In this challenge, participants are tasked with predicting the production of rice on a state-wise or union territory-wise basis. The dataset provided spans from the agricultural sessions of 2004-2005 to 2022-2023, detailing the quantity of rice produced annually.
### Proposal Overview
The prediction of rice production on a state-wise or union territory-wise basis is vital for agricultural planning and policymaking. In this proposal, we outline our approach to analyze the dataset spanning from 2004-2005 to 2022-2023 and predict rice production using two different models: ARIMA (AutoRegressive Integrated Moving Average) and LSTM (Long Short-Term Memory). Our goal is to derive meaningful insights from the data and provide accurate predictions for the next five years.
### Dataset Analysis
#### Overview
The dataset comprises three distinct tables, each offering insightful perspectives on rice production spanning from 2004-2005 to 2022-2023. Within Table 1 and Table 2, comprehensive state-wise data delineates the intricate nuances of rice production dynamics over the years.
#### Analysis Goals
Our analysis delves deeper to determine the rate of production growth or decline for each state/union territory. By scrutinizing the annual production trends against historical data, we can discern not only the magnitude but also the trajectory of change in rice production across different regions.
#### Data Visualization
We utilize bar charts, pie charts, and line plots to comprehensively analyze state-wise production variations over the years.
### ARIMA Model
#### Preprocessing
We ensure the reliability of our analysis by prioritizing the attainment of stationarity within the time series data.
#### Model Selection
We employ the ARIMA model and conduct hyperparameter tuning to achieve the best possible fit to the data.
#### Training and Evaluation
We train the ARIMA model using historical data and subject it to rigorous evaluation using metrics such as MAE and RMSE.
#### Prediction
We make predictions for the next five years using the trained ARIMA model.
### LSTM Model
#### Data Preparation
We transform the dataset into sequences of fixed length to prepare it for training the LSTM model.
#### Model Architecture
We carefully design an LSTM architecture considering factors such as the number of layers, units, and dropout rates.
#### Training and Validation
We train the LSTM model using historical data and validate its performance on a separate test set.
#### Prediction
We make predictions for the next five years using the trained LSTM model.
## Comparative Analysis
We conduct a comprehensive comparative analysis of the ARIMA and LSTM models for forecasting state-wise rice production.
## Conclusion
Our proposed approach aims to provide accurate forecasts for agricultural planning and decision-making, contributing to improving rice production strategies and addressing challenges in agricultural sustainability.
---
# Developing a Multimodal Model for Detecting Harmful Internet Memes using CLIP + CNET Architecture
## Introduction
The proliferation of internet memes, often combining images and text, has become a significant cultural phenomenon. In this proposal, we outline our approach to develop a novel multimodal machine learning model for classifying harmful internet memes using the CLIP (Contrastive Language-Image Pre-training) model in conjunction with a custom convolutional neural network (CNet) architecture.
## Dataset Description
We utilize the "BrainDead Multimodal Data for Hateful Meme" dataset provided for the competition, consisting of 10,000 data points, each representing an internet meme.
## Proposed Approach
We outline our approach, including preprocessing, CLIP model integration, CNet architecture design, fusion, and classification.
## Comparative Analysis
We compare the performance of our proposed model with similar datasets A and B, utilizing performance metrics and insights gained from the comparison.
## Conclusion
Our proposed multimodal model aims to achieve superior performance compared to existing benchmark models while minimizing computation power requirements, contributing to the advancement of meme moderation systems.
---
## Credits
- Vanshika Kothari
- Sayan Roy
- Harsh Raj Gupta
- Harsh Ranjan
- Aaratrika Sarkar
We extend our sincere gratitude to all team members for their exceptional contributions to our ML model report. Each team member's dedication, expertise, and collaborative spirit were truly commendable.
## References