Fast Incremental Learning With Swarm Decision Table and Stochastic Feature Selection in an IoT Extreme Automation Environment
Tengyue Li, Simon James Fong, Richard Charles Millham, Jinan Fiaidhi, Sabah Mohammed · IT Professional · 2019
Fog computing and a collection of incremental learning algorithms are introduced to overcome the challenge of enabling smart home applications. Specifically, there are two cases of Internet-of-things (IoT) sensing extreme automation (EA) scenarios, i.e., one being autorecognizing the gas type from the air composition and the other predicting the next unusual energy consumption, are looked into. Our experimentation over a collection of the popular incremental learning algorithm, such as HT and our proposed novel SDT, as well as heuristics and metaheuristics feature selection methods, are presented. The simulation results show that compared with Hoeffding tree (HT), SDT is more stable in performance. HT is fast because the model representation is in a tree format; tree traversal is fast by design. SDT learns and modifies the modeling rules in the style of k-nearest neighbor, checking on many data points within the decision table in each update. The paper demonstrates a simulation experiment combining a feature selection method coupled with a swarm intelligence and decision Table classifier as SDT to find the most suitable SDT models in the Fog computing environment. According to the results, we find three appropriate swarm feature selection algorithms: BestFirst, Elephant, and Harmony. In addition, each SDT model demonstrates its unique ability in terms of recovery ability, fluctuation e degree, and successive time.