This paper implements deep learning methods of recurrent neural networks and short-term memory models. Two kinds of time-series data were used: air pollutant factors, such as O₃, SO₂, and CO₂ from 2017 to 2019, and meteorological factors such as temperature, humidity, wind direction, and wind speed. A trained model was used to predict air pollution within an eight-hour period. Correlation analysis was applied using Pearson and Spearman correlation coefficients. The KNN method was used to fill in the missing values to improve the generated model’s accuracy. The average absolute error percentage value was used in the experiments to evaluate the model’s performance. LSTM had the lowest RMSE value at 1.9 than the other models from the experiments. CNN had a significant RMSE value at 3.5, followed by Bi-LSTM at 2.5 and Bi-GRU at 2.7. In comparison, the RNN was slightly higher than LSTM at a 2.4 value.

In order to achieve the Sustainable Development Goals (SDG), it is imperative to ensure the safety of drinking water. The characteristics of each drinkable water, encompassing taste, aroma, and appearance, are unique. Inadequate water infrastructure and treatment can affect these features and may also threaten public health. This study utilizes the Internet of Things (IoT) in developing a monitoring system, particularly for water quality, to reduce the risk of contracting diseases. Water quality components data, such as water temperature, alkalinity or acidity, and contaminants, were obtained through a series of linked sensors. An Arduino microcontroller board acquired all the data and the Narrow Band-IoT (NB-IoT) transmitted them to the web server. Due to limited human resources to observe the water quality physically, the monitoring was complemented by real-time notifications alerts via a telephone text messaging application. The water quality data were monitored using Grafana in web mode, and the binary classifiers of machine learning techniques were applied to predict whether the water was drinkable or not based on the data collected, which were stored in a database. The non-decision tree, as well as the decision tree, were evaluated based on the improvements of the artificial intelligence framework. With a ratio of 60% for data training: at 20% for data validation, and 10% for data testing, the performance of the decision tree (DT) model was more prominent in comparison with the Gradient Boosting (GB), Random Forest (RF), Neural Network (NN), and Support Vector Machine (SVM) modeling approaches. Through the monitoring and prediction of results, the authorities can sample the water sources every two weeks.

Optimization strategies in deep learning models require different techniques for different use cases. Besides, various phases of the model deployment life-cycle specify possible and particular optimization strategies. In this paper, an optimized deep learning model on the edge computing environment is proposed for image classification cases. For preparing the dataset, the image preprocessing and data augmentation methods are utilized to prepare the data for the training process. To accelerate the deep learning training process, this system implemented CPU optimization and hyperparameter tuning. Tensorflow is applied as a framework for the training model. InceptionV3, VGG16, and MobileNet are applied as topology implemented in the deep learning training comparison. In this case, InceptionV3 was used for modeling the deep learning applications on edge. To optimize the trained model, a Model Optimizer is used on the edge device. It can be seen in the experiments, MobileNet was the least accurate model (85%) and the longest time to load the model (71s). VGG16 was the most reliable (91%) and the shortest time to load the model (50s). InceptionV3 has median accuracy (87%) and the average time to load the model (52s).