🍃Air_Quality_Analysis_and_Prediction_for_Indian_Cities_and_States

This project aims to analyze and predict the air quality index of various cities and states in India using machine learning. The dataset used in this project contains historical data of air quality index readings from various monitoring stations across India.

💡 inspiration and What is unique in the project

in my city indore there are several places that show aqi and chemical levels troughout the day in a tabular format, i researched about this topic and found out this research paper https://www.diva-portal.org/smash/get/diva2:1681590/FULLTEXT02 in this research paper they have just taken a small amount of data and shoved it into models

• in this project i have tried my best to make and better model with more data 100000, compared to 30000 in the research paper
• and extract usefull and meaningfull insights about aqi trend in india using interactive and good looking maps

YOU CAN CHECK THE READMES FOR PART 1 AND 2_3 OF THIS PROJECT AND CHECK DEMO VIDEO FOR THE OUTPUT

📚 PARTS OF THE PROJECT

The project is divided into 3 Parts:

📙 1_exploratory_analysis_of_aqi_in_india.ipynb

• Preliminary and heavy Exploratory analysis of historical aqi data from indian states and citites has been done in this

• main visual insights are as follows:

  1. trend of aqi throughout the years
  2. aqi trend of states in india
  3. aqi trend of cities in india
  4. yearly/ monthly state_Aqi_from_2015_to_2020 function
  5. yearly/ monthly city_Aqi_from_2015_to_2020 function
  6. correlation of toxic gasses with aqi (top 3)
  7. yearly/monthly state_toxic_gas_from_2015_2020 india function
  8. yearly/monthly city_toxic_gas_from_2015_2020 india function
  9. aqi of states on an india map
  10. aqi of cities on an india map

an example of these visualization, 10. aqi of cities on an india map

refer to 1_exploratory_analysis_of_aqi_in_india.ipynb or README_PART1 to get insights on more visualizations

🛃 staion_merged_csv is made in this ipynb by combing station csv and station_day csv

📙 2_model_analysis_of_aqi_in_india.ipynb

this uses the station_merged_csv and does performs these tasks:

• Exploratory analysis of Columns has been performed and heavy data preprocessing has been done

• then this preprocessed data is used to train 3 models: multiple linear reggresion, SVR, XGBRegressor

• the best model made was xgb_model (Training score: 0.95, Testing score: 0.91, MEA : 21.22 , MSE: 1370.24, RMSE: 37)

🛃 xgb_model was made in this ipynb

📙3_main_AQI_predictor.pyw

this is the main gui program that uses the xgb_model to predict aqi of a region based on 12 factors and show the output to the user

what readmes to read first:

• for more insights and detailed process regaurding exploratory and preliminary analysis with visual and interacvtible graphs refer README_PART1 (part 1 of this project)

• For more insights on data distribution, data preprocessing and inputs/ outputs of 3_main_AQI_predictor refer to README_PART_2_3

• 🛃 in order README --> README_PART1 --> README_PART_2_3

📚 about the dataset:

• the dataset station_merged csv is a combination of station_day csv and station csv

in station_merged_data / staion_merged_csv/ staion_merged_df

#	Column	Non-Null Count	Dtype	meaning
0	StationId	108035 non-null	object	id of the station
1	City	108035 non-null	object	city in which the station is located
2	State	108035 non-null	object	state in which the station is located
_	Date	108035 non-null	object	date of the readings (now removed)
3	PM2.5	86410 non-null	float64	amount of said checmical in the air
4	PM10	65329 non-null	float64	amount of said checmical in the air
5	NO	90929 non-null	float64	amount of said checmical in the air
6	NO2	91488 non-null	float64	amount of said checmical in the air
7	NOx	92535 non-null	float64	amount of said checmical in the air
8	NH3	59930 non-null	float64	amount of said checmical in the air
9	CO	95037 non-null	float64	amount of said checmical in the air
10	SO2	82831 non-null	float64	amount of said checmical in the air
11	O3	82467 non-null	float64	amount of said checmical in the air
12	Benzene	76580 non-null	float64	amount of said checmical in the air
13	Toluene	69333 non-null	float64	amount of said checmical in the air
14	Xylene	22898 non-null	float64	amount of said checmical in the air
15	AQI	87025 non-null	float64	the air quality that day
16	AQI_Bucket	87025 non-null	object	AQI catagory ranges from good to severe
17	year	87025 non-null	float64	year of the reading
18	month	87025 non-null	float64	month of the reading
18	date	87025 non-null	float64	day of the reading

🪟 where you can get this data?

The dataset is avaible on kaggle this is the link: https://www.kaggle.com/datasets/rohanrao/air-quality-data-in-india

• city_day csv
• city_hour csv
• station_day csv
• station_hour csv
• station csv

The links for geo json files of india are as follows:

• state geojson: https://github.com/Subhash9325/GeoJson-Data-of-Indian-States
• district geojson: https://www.kaggle.com/datasets/krutarthhd/india-geojson-file?select=district

🔚 Conclusion

many insights have been made in part 1 of this project some of which are as follows:

• the AQI of india has steadily decreaseed over the years the present having the freshest air

• Gujrat has the worst aqi followed by delhi when it comes to indian states, mizoram has the best aqi

• ahemdabad has the worst aqi followed by delhi when it comes to indian citites aizawl has the best aqi

• PM2.5, PM10, NO2 are the biggest contributers of air polution and aqi

• many many more have been done refer to README_PART1.md for more insights

after analysing the data and training the model these insights have been extracted

• The best model in the research paper was (Ridge regression mea = 27.907, rmse = 36.791, r2 = 0.8089) on city_data csv

• my best model in this notebook was (xgb_model mea = 21.22, rmse = 37.011, r2 = 0.91) on station_data csv

• the city_data csv has only 30000 rows while station_data csv has 100000 which is way more data

• my model has achived better mea and r2 scores on a bigger scale of data making it better than the ridge reggression model made in the research paper

• but my is not the best and can still be upgraded

• xgb_model has worse rmse of 37.011 which is 1 units more than rigde regression model

⛔ Limitations

• the project is available onlyon pc with python installed

🔜 future upgrades for the project

• making the model avaible on mobile apps
• a dedicated website for the model
• a better model which increases the rmse score

👨‍🦱 credits and contact info

• made by Aryan Rathore
• LinkedIn : https://www.linkedin.com/in/aryan-rathore-b15459215/
• email: aryanrathore13572002@gmail.com, aryan.rathore2021@vitbhopal.ac.in

----------------------------------------------------------------