Use of Socio-economic, Climatic, and Land use Land Cover Patterns in Solid Waste Forecasting with Integrated Gradient LSTNet Based Model in Lomé, Togo

Figures & data

Figure 1. Multivariate time series dataset.

Figure 2. Pearson correlation coefficient of features.

Figure 3. Features selection from GBR and RFR.

Figure 4. Models forecasting results on the test dataset.

Table 1. Models parameters and performances.

Model	Hyperparameters				Forecasting
	Layer	Optimizer	Activation function	Dropout	RSE	RAE
LSTM	2-layer LSTM (64, 32)	SGD	tanh	0.2	1.0253	1.0708
CNN	1D-CNN (32)	SGD	tanh	0.2	1.1080	1.1727
CNN+LSTM	1D-CNN (32), 1-layer LSTM (32)	SGD	tanh	0.2	1.0134	1.0595
LSTNet+IG	1D-CNN, GRU, GRUskip, Linear	Adam	Sigmoid	0.3	0.9727	0.9108
RFR+LSTNet+IG	1D-CNN, GRU, GRUskip, Linear	Adam	Sigmoid	0.3	0.9782	0.9900
GBR+LSTNet+IG	1D-CNN, GRU, GRUskip, Linear	Adam	Sigmoid	0.3	1.1873	1.1903
RFR+GBR+LSTNet+IG	1D-CNN, GRU, GRUskip, Linear	Adam	Sigmoid	0.3	1.0145	1.0892

Figure 5. LSTNet+IG features influence intensity.

Figure 6. (a) RFR+LSTNet+IG features influence, (b) GBR+LSTNet+IG features influence, and (c) RFR+GBR+LSTNet+IG features influence.

Data Availability Statement

Data is made available publicly.