Forecasting of Traffic Flow Using Feature Selection with ML Model

Intelligent traffic control and public risk assessment are only two examples of the many spatial-temporal mining applications that rely on accurate predictions of citywide traffic flow. The ability to select an optimal traffic variable grouping is crucial to traffic flow prognostication, as it can cut down on calculation time and boost accuracy. Although the linear traffic system problem can be easily solved using the conventional method, its inability to adapt to sudden shifts in traffic makes it challenging to use. The intelligent forecasting approach works well for managing short-term traffic flow and can realise arbitrary non-linear mapping. Unfortunately, its effectiveness is highly sensitive to the supplied data. Parameter selection is another challenging topic that necessitates a big training set. To this end, this work proposes a feature selection method employing a hybrid optimization model (the slime procedure (SMA) and the whale optimization-(WOA)). To begin, the suggested model assesses the importance and uniqueness of each variable, and then uses feature selection to pick the most important ones and eliminate the duplicates. Secondly, the forecasting engine is Extreme Machine Learning (ELM). Lastly, the suggested technique's forecasting performance is validated using a numerical instance using traffic flow data from Pems; the consequences show that the proposed method may successfully lower the cost of control while simultaneously increasing the model's prognostication precision.