human-doodle/air-pollution-prediction

Time series prediction using LSTM

Air Pollution Prediction : Project Overview

• Forcasted the pollution at the next hour given the weather conditions and pollution for prior hours.
• Worked on the dataset from UCI Machine Learning Repository that reports on the weather and level of pollution each year at the US Embassy in Beijing
• Prepared data and transformed it into a supervised learning problem.
• Fitted an LSTM on the multivariate- input data.

Data Cleaning & Preparation

Made following changes to make it usable for the model:

• Consolidated the date-time information into a single date-time to use as an index in Pandas.
• "No" column dropped, rows with na values dropped and columns renamed.
• Normalised the input variables.
• Transformed the existing dataset into a supervised learning problem to predict the pollution at current hour given pollution measurement and weather conditions at the prior timestep.
• The weather variables for the hour to be predicted are removed
• One-hot encoded the wind-dir

EDA

Looked at the distributions of the data.

Model Building

• The data was split into train and test set with 4 years of data in training set and 1 year of data in test set.

• Then the sets were split into input(X) and output(y) variables

• Reshaped the inputs(X) into 3D format as expected by LSTMs - [samples, timestamps, features]

• Defined LSTM with 50 neurons in the first hidden layer and 1 neuron in the output layer for prediction.

• Used Mean Absolute Error(MAE) loss function and the Adam version of stocastic gradient descent.

• Fit the model for 50 training epochs with a batch size of 72.

• Plotted graph for training and test losses

• 

Model performance

• Model acheived Root Mean Squared Error(RMSE) of 25.004

Code and Resources Used

Python Version: 3.10.4
Packages: keras with tensorflow backend, SciPy, Scikit-learn, pandas, numPy, matplotlib Reference: Machine Learning Mastery- Jason Brownlee