District-HeatIndex-Predication-using-LSTM

Model Compare Table:

Model	Feature No.	Test Loss	Test Accuracy	Predicted HEAT_INDEX	Actual HEAT_INDEX	Mean Absolute Error
1	8	1.52	90.45	73.28	74.71	1.43%
2	4	1.50	90.16	74.79	74.71	0.08%
3	25	0.34	40.56	73.9	74.71	0.91%

model 2 shows best performance as its Mean Absolute Error is very less than Model 1 and Model 3...

Features for each Model:

Model 1 : ['T2M', 'T_Fahrenheit', 'TS', 'T2M_MIN', 'T2M_MAX', 'T2MWET', 'WS10M_RANGE', 'DISTRICT_INDEX']
Model 2 : ['T2M','RH2M', 'T2MDEW', 'WD10M']
Model 3 : ['T2M', 'T_Fahrenheit', 'TS', 'T2M_MIN', 'T2M_MAX', 'T2MWET', 'WS10M_RANGE', 'WD50M', 'WD10M', 'WS10M_MAX', 'RH2M', 'MO', 'WS10M', 'T2MDEW', 'QV2M', 'WS50M', 'LON', 'DISTRICT_INDEX', 'LAT', 'PS', 'PRECTOTCORR', 'YEAR', 'T2M_RANGE', 'DY', 'WS10M_MIN']

Project Overview:

Title:

HeatIndex Prediction using LSTM

Description:

LSTM models predict Heat Index with varying feature sets. Model 2 excels, showcasing superior performance with fewer features.

Problem Statement:

Predict Heat Index based on meteorological features for improved understanding of weather conditions and human comfort.

Abstract:

This project utilizes LSTM models to predict Heat Index, comparing performance with different feature sets. Model 2, with 4 selected features, outperforms others.

Application:

Weather Forecasting
Urban Planning
Health and Safety
Agriculture
Energy Consumption
Outdoor Event Planning

Advantages:

Accurate Heat Index Prediction
Reduced Model Complexity (Model 2)
Better Interpretability (Model 2)
Faster Training and Inference (Model 2)
Improved Resource Efficiency

Limitations:

Limited Feature Set
Potential Overfitting
Sensitivity to Input Data Quality
Lack of External Factors
Model Generalization Challenges

Future Scope:

Include Additional Relevant Features
Explore Advanced LSTM Architectures
Fine-Tune Hyperparameters
Integrate External Data Sources
Develop Real-time Prediction Capability

About Dataset:

Features:

T2M : Temperature at 2 meters (in degrees Celsius).
T_Fahrenheit : Temperature at 2 meters converted to Fahrenheit.
TS : Surface temperature (in degrees Celsius).
T2M_MIN : Minimum temperature at 2 meters (in degrees Celsius).
T2M_MAX : Maximum temperature at 2 meters (in degrees Celsius).
T2MWET : Wet-bulb temperature at 2 meters (in degrees Celsius).
WS10M_RANGE : Wind speed at 10 meters range (difference between maximum and minimum) (in meters per second).
WD50M : Wind direction at 50 meters (in degrees).
WD10M : Wind direction at 10 meters (in degrees).
WS10M_MAX : Maximum wind speed at 10 meters (in meters per second).
RH2M : Relative humidity at 2 meters (percentage).
MO : Month.
WS10M : Wind speed at 10 meters (in meters per second).
T2MDEW : Dew point temperature at 2 meters (in degrees Celsius).
QV2M : Specific humidity at 2 meters (in grams per kilogram).
WS50M : Wind speed at 50 meters (in meters per second).
LON : Longitude.
DISTRICT_INDEX : Numerical index representing the district.
LAT : Latitude.
PS : Surface pressure (in pascals).
PRECTOTCORR : Corrected total precipitation (in millimeters).
YEAR : Year.
T2M_RANGE : Temperature range at 2 meters (in degrees Celsius).
DY : Day.
WS10M_MIN : Minimum wind speed at 10 meters (in meters per second).

Dataset Information:

Rows: 125,060
Columns: 25
Memory Usage: 23.9 MB

About LSTM Algo:

Explaination

LSTM (Long Short-Term Memory) is a type of recurrent neural network (RNN) designed for sequence prediction. In our project, LSTM is applied to time-series meteorological data to learn patterns and predict heat index.

Application

Meteorological Forecasting
Public Health Planning
Urban Planning for Heat Resilience
Agriculture Heat Stress Monitoring
Energy Consumption Forecasting
Outdoor Event Planning

Advantages

Captures Long-Term Dependencies
Handles Time-Series Data Effectively
Suitable for Complex Patterns
Avoids Vanishing Gradient Problem
Allows End-to-End Learning

Limitations

Requires Sufficient Data
May Overfit with Limited Data
Computationally Intensive
Hyperparameter Sensitivity
Limited Interpretability

Future Scope

Enhance Model Robustness
Incorporate Additional Features
Explore Ensemble Approaches
Integrate Real-Time Data
Extend to Multiple Locations

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

Model 1: (8 Feature)

selected_features : ['T2M', 'T_Fahrenheit', 'TS', 'T2M_MIN', 'T2M_MAX', 'T2MWET', 'WS10M_RANGE', 'DISTRICT_INDEX']
input_data : pd.DataFrame({ 'T2M': [0.325153], 'T_Fahrenheit': [72.91], 'TS': [0.336126], 'T2M_MIN': [0.431715], 'T2M_MAX': [0.305014], 'T2MWET': [0.631997], 'WS10M_RANGE': [0.164191], 'DISTRICT_INDEX': [10], })
Output Predicted HEAT_INDEX: 73.28 Actual HEAT_INDEX: 74.71 Mean Absolute Error: 1.430001 %

Model 2: (4 Feature)

selected_features : ['T2M','RH2M', 'T2MDEW', 'WD10M']
input_data : pd.DataFrame({ 'T2M': [0.325153], 'RH2M': [0.733267], 'T2MDEW': [0.766538], 'WD10M': [0.614839], })
Output Predicted HEAT_INDEX: 74.79 Actual HEAT_INDEX: 74.71 Mean Absolute Error: 0.080001 %

Model 3: (25 Feature)

selected_features : ['T2M', 'T_Fahrenheit', 'TS', 'T2M_MIN', 'T2M_MAX', 'T2MWET', 'WS10M_RANGE', 'WD50M', 'WD10M', 'WS10M_MAX', 'RH2M', 'MO', 'WS10M', 'T2MDEW', 'QV2M', 'WS50M', 'LON', 'DISTRICT_INDEX', 'LAT', 'PS', 'PRECTOTCORR', 'YEAR', 'T2M_RANGE', 'DY', 'WS10M_MIN']
input_data : pd.DataFrame({ 'T2M' : [0.325153], 'T_Fahrenheit' : [72.91], 'TS' : [0.336126], 'T2M_MIN' : [0.431715], 'T2M_MAX' : [0.305014], 'T2MWET' : [0.631997], 'WS10M_RANGE' : [0.164191], 'WD50M' : [0.609584], 'WD10M' : [0.614839], 'WS10M_MAX' : [0.175806], 'RH2M' : [0.733267], 'MO' : [1], 'WS10M' : [0.176881], 'T2MDEW' : [0.766538], 'QV2M' : [0.549058], 'WS50M' : [0.208062], 'LON' : [79.293], 'DISTRICT_INDEX' : [10], 'LAT' : [19.366], 'PS' : [0.43377], 'PRECTOTCORR' : [0.031997], 'YEAR' : [2001], 'T2M_RANGE' : [0.367852], 'DY' : [1], 'WS10M_MIN' : [0.139295], })
Output Predicted HEAT_INDEX: 73.8 Actual HEAT_INDEX: 74.71 Mean Absolute Error: 0.909997 %

AtharvaPawar456 / District-HeatIndex-Predication-using-LSTM