AIN311 Machine Learning in Sustainability Project: Week 4 — Unveiling the Power of 1DCNN
Welcome back sustainability enthusiasts! Week 4 of our Machine Learning in Sustainability project is here and we’re feeling a bit experimental this week. Therefore, we are diving deep into the realms of 1D Convolutional Neural Networks (1DCNN). Hold onto your bikes as we explore the cutting edge technology that promises to take our metrics to new heights.
🌿 1DCNN: Reshaping the Landscape
As Linear Regression and Random Forest Regression are done, our new participant, the 1DCNN, takes the spotlight. Convolutional Neural Networks renowned for their prowess in image recognition and have been adapted to handle the unique challenges posed by 1D data such as time series and sequence data common in sustainability metrics.
📊 Layers Showdown: Unveiling Patterns
We arranged our time series data to the shape of 24 to use our model. Then we head right onto the our layers
Our 1DCNN has a deep personality with 6 layers. 2 of them are convolution layers, 1 pooling, 1 flatten and 2 fully connected. Convolution layer 1 seeks the fundamental features we have to identify the essential building blocks of our data while convolution layer 2 dives even deeper into our futures to grasp more and complex patterns in our data.
Then comes the pooling layer, it reduces our dimensions and by reducing spatial dimensions it ensures that our model keeps the crucial insights while helping the computational process.
Next to flattennig layer, it projects our output into 1D to prepare our dimensions for the next step.
Lastly our fully connected layers, The first fully connected layer (fc1) functions as the global thinker. It looks at our output of patterns discovered by the convolutional layers, having nonlinearity with ReLU activation to capture the patterns of our sustainability landscape. Second fully connected layer serves as the synthesizer, putting all learned insights into a final output of size 3. It's our decision maker, which gives us the final understanding of sustainability metrics.
🔄 Results Showdown: The Final
Here comes our results showdown for the week! We had great results this week with following results
We have obtained so little normal and validation loss which shows our model is quite successful. The subject of overfitting is on the table for further examination. We are still feeling experimental so what’s going to happen next week will be a suprise but it is guaranteed to be fun! See you on next week!
